• Supply Chain Sanctions: The Pentagon is moving to designate Anthropic as a "supply chain risk" due to the company's refusal to grant unrestricted military access to its AI model, Claude.
• Policy Deadlock: Negotiations over the renewal of the "Claude Gov" classified version have stalled as Anthropic seeks to prohibit use for autonomous weaponry and mass surveillance of American citizens.
• Mandatory Divestment: A formal risk designation would require all defense contractors and their suppliers to remove Claude from their internal workflows and document analysis systems.
• Venezuela Operation: Reports indicate that Claude was utilized through a Palantir partnership during a January military raid in Caracas, despite company terms prohibiting violent deployments.
• Operational Demands: Defense officials are demanding an "all lawful purposes" standard for AI usage, arguing that ethical guardrails create unworkable gray areas for battlefield commanders.
• Market Competition: Competitors including Google, OpenAI, and xAI have reportedly agreed to remove military guardrails on unclassified systems and are seeking access to classified networks.
• Technical Superiority: Claude Gov currently remains the only AI model operating on the military's classified networks, as competing models are regarded by some officials as less advanced for intelligence analysis.
• Economic Leverage: The Pentagon is using the supply chain designation as a tool to signal that domestic technology firms must prioritize military requirements over internal ethical frameworks to maintain government eligibility.

The Pentagon is preparing to designate Anthropic as a "supply chain risk" and sever business ties with the AI company over its refusal to allow unrestricted military use of Claude, Axios reported on Monday. Defense Secretary Pete Hegseth is "close" to making the decision after months of failed negotiations, a senior Pentagon official told the outlet. "It will be an enormous pain in the ass to disentangle, and we are going to make sure they pay a price for forcing our hand like this," the official said.

Supply chain risk designations are typically reserved for foreign adversaries and hostile actors, the kind of label slapped on Huawei routers or Russian software vendors. Not American companies. Applying it to Anthropic, widely regarded as one of the country's leading AI developers, would represent an extraordinary act of retaliation by the Pentagon against a domestic technology firm.

The Breakdown

• Pentagon preparing to label Anthropic a "supply chain risk," forcing all military contractors to drop Claude
• Dispute centers on mass surveillance and autonomous weapons; Pentagon demands "all lawful purposes" standard
• Claude reportedly used in Venezuela/Maduro raid via Palantir; Anthropic's terms prohibit violent deployment
• Google, OpenAI, and xAI agreed to remove military guardrails but none yet operate on classified networks

What the fight is about

Anthropic and the Pentagon have spent months in contentious negotiations over the renewal of Claude Gov, the classified version of Claude built specifically for the U.S. national security apparatus. Two issues are holding up the deal. Anthropic wants to ensure Claude isn't used for mass surveillance of American citizens or to develop weapons that fire without human involvement. The Pentagon is demanding the right to use Claude for "all lawful purposes," a standard it is also pressing on Google, OpenAI, and Elon Musk's xAI.

"The Department of War's relationship with Anthropic is being reviewed," chief Pentagon spokesperson Sean Parnell told The Hill on Monday. "Our nation requires that our partners be willing to help our warfighters win in any fight. Ultimately, this is about our troops and the safety of the American people."

The company's position is that current U.S. surveillance law was never written with AI in mind. The government already collects enormous quantities of personal data, from social media posts to concealed carry permits. AI could supercharge that authority in ways existing statutes don't contemplate. Think pattern-matching across millions of civilian records, cross-referencing behavior at a scale no human analyst corps could replicate. Anthropic has said it is prepared to loosen its current terms but wants explicit boundaries.

Pentagon officials counter that the restrictions are too broad and would create unworkable gray areas on the battlefield. Defense officials have argued that drawing bright lines around specific capabilities would hamper operations in scenarios nobody can fully predict yet, a concern that Anthropic's critics inside the Pentagon call naive at best and obstructionist at worst.

The relationship appears to have soured beyond the specifics of the contract. A source familiar with the negotiations told Axios that senior defense officials had been frustrated with Anthropic for some time and "embraced the opportunity to pick a public fight."

The Venezuela trigger

The Wall Street Journal reported last week that Claude was used during the U.S. military operation in January that captured Venezuelan President Nicolas Maduro from his residence in Caracas. The AI tool reached the battlefield through Anthropic's partnership with Palantir Technologies, which holds extensive military and intelligence contracts. That revelation turned a simmering contract dispute into something more volatile.

Venezuelan authorities said 83 people were killed in bombings across the capital during the raid. Anthropic's own terms of use prohibit Claude's deployment for violent purposes, weapons development, or surveillance. Claude Gov has capabilities ranging from processing classified documents and intelligence analysis to cybersecurity data interpretation, and it was unclear which of those functions the military called on during the operation.

Anthropic declined to comment on whether its technology played a role but said any use would need to comply with its usage policies. Palantir and the Pentagon declined to comment.

Reports of the raid prompted Anthropic to inquire about whether its technology had been involved, according to The Hill, though the company denied making any outreach to the Pentagon or Palantir about the incident. The question itself was enough to deepen the rift. Here was a company that had staked its brand on responsible development, now staring at reports that its technology may have supported a military operation that killed dozens of people. Nobody confirmed anything. Nobody denied it either. And somewhere in Arlington, the contract renewal was still sitting on a desk.

What supply chain risk actually means

If the Pentagon goes through with the designation, the consequences reach far beyond Anthropic's military contract. Every company that does business with the Pentagon would need to certify that it does not use Claude in its own workflows. That's the kind of compliance burden normally imposed over Chinese telecom equipment or adversary-state technology. Not a San Francisco AI startup.

Losing the military deal itself wouldn't cripple Anthropic. The two-year contract, announced last July, is worth up to $200 million, a fraction of the company's reported $14 billion in annual revenue. The real damage is the ripple.

Stay ahead of the curve

Strategic AI news from San Francisco. No hype, no "AI will change everything" throat clearing. Just what moved, who won, and why it matters. Daily at 6am PST.

No spam. Unsubscribe anytime.

Eight of the ten largest U.S. companies use Claude, according to Anthropic. The technology sits inside enterprise systems across finance, healthcare, legal services, and government, running on laptops in procurement offices and servers in contractor data centers. Defense contractors and their suppliers who rely on Claude for document analysis, coding assistance, or internal operations would all need to strip it out or prove they'd stopped. That's the "enormous pain in the ass" the Pentagon official was describing, and it lands on the companies that depend on Anthropic's tools, not just on Anthropic itself.

Then there's the market signal. If you're an enterprise buyer evaluating AI vendors, the designation forces a new question into the procurement conversation: does choosing Claude put future government contracts at risk? That kind of uncertainty doesn't hit revenue immediately. It shows up in renewal conversations, in procurement meetings where a general counsel raises a flag, in RFP language that quietly specifies "Pentagon-compatible AI providers." The Pentagon knows this. A $200 million contract is a bargaining chip. A supply chain designation is a weapon.

Competitors already in the room

Pentagon officials have made clear they have alternatives. Google, OpenAI, and xAI have all agreed to remove their guardrails for military use on unclassified systems, Axios reported. All three are negotiating access to classified military networks. Pentagon officials expressed confidence that the companies would accept the "all lawful purposes" standard.

But swapping out Anthropic would not be painless. A senior administration official acknowledged that competing models "are just behind" for specialized government applications. Claude Gov was purpose-built for handling classified materials, interpreting intelligence, and processing cybersecurity data. It remains the only AI model running on the military's classified networks. OpenAI has built a custom GenAI.mil tool for the Pentagon and other allied nations. Google already provides a customized version of Gemini for defense research. xAI, owned by Musk, signed a Pentagon agreement in January. None of them operate on classified systems yet.

Officials who have used Claude Gov rate it highly. No complaints about the tool's capability during the Maduro operation or anywhere else. The grievance is entirely about terms, not technology. Which makes the threatened divorce harder to read at face value: the Pentagon wants to banish the product it praises most.

Precedent matters as much as the switch itself. How Anthropic's negotiations resolve will shape the contract terms for every AI company seeking classified military work. OpenAI, Google, and xAI are watching. A source familiar with those discussions told Axios that "much is still undecided" despite the Pentagon's confidence, suggesting the three companies haven't fully committed to whatever the Pentagon is asking behind closed doors.

A loyalty test dressed as policy

Anthropic has built its identity around the idea that powerful AI requires guardrails. CEO Dario Amodei has called publicly for regulation to prevent catastrophic harms from AI. He has warned against autonomous lethal operations and mass surveillance on U.S. soil. The company raised billions in funding partly on the premise that responsible development and commercial success are compatible goals. Investors bought that vision. Enterprise customers bought it. The Pentagon, for a while, bought it too.

Hegseth tested that premise in January when he told reporters the Pentagon wouldn't "employ AI models that won't allow you to fight wars." That framing collapses Anthropic's specific objections, on surveillance law, on weapons autonomy, into something blunter: a refusal to cooperate. It turns a policy dispute into a question of allegiance. Other AI companies appear to have taken note. Google employees staged walkouts over a Pentagon drone contract called Project Maven back in 2018. That was a different era. Google is now negotiating classified network access for Gemini, no Anthropic-style restrictions attached.

Anthropic hasn't walked away from the table. A spokesperson told The Hill on Monday that Anthropic remained "committed" to supporting U.S. national security and described the ongoing conversations as "productive" and conducted "in good faith." The company pointed out it was the first frontier AI developer to put models on classified networks and the first to build customized models for national security customers.

Good faith has limited currency when the other side is telling reporters you'll "pay a price." Pentagon officials aren't characterizing this as a disagreement between reasonable parties hashing out contract language. They're treating Anthropic's ethical boundaries like a contamination risk, something to be flagged and cut out of the supply chain entirely.

Anthropic built the most advanced classified AI system the Pentagon has ever deployed. That same technology reportedly played a role in an active military raid. Now the company faces classification alongside hostile foreign suppliers, not because the technology failed, but because its maker wanted a say in how it gets used. The contamination the Pentagon identified isn't in the code. It's in the conditions.

Frequently Asked Questions

What is a supply chain risk designation?

A Pentagon classification that bars a company's products from the U.S. military supply chain. It's typically reserved for foreign adversaries like Chinese telecom firms. If applied to Anthropic, every defense contractor would need to certify it doesn't use Claude, potentially disrupting operations at companies across finance, healthcare, and government.

What is Claude Gov?

A specialized version of Anthropic's Claude AI built for the U.S. national security apparatus. It handles classified materials, interprets intelligence, and processes cybersecurity data. It is currently the only AI model running on the military's classified networks, giving Anthropic a unique position among AI companies serving the Pentagon.

How was Claude used in the Venezuela operation?

The Wall Street Journal reported that Claude was deployed during the January military operation that captured Venezuelan President Nicolas Maduro, accessed through Anthropic's partnership with Palantir Technologies. The specific role Claude played remains unclear. Anthropic declined to confirm involvement, and both Palantir and the Pentagon declined to comment.

What restrictions is Anthropic insisting on?

Anthropic wants contract language preventing Claude from being used for mass surveillance of American citizens or to develop weapons that fire without human involvement. The company argues current surveillance law doesn't account for AI's ability to cross-reference civilian data at massive scale. The Pentagon considers these restrictions too broad for military operations.

Are other AI companies willing to work with the Pentagon without restrictions?

Google, OpenAI, and xAI have agreed to remove guardrails for military use on unclassified systems and are negotiating classified network access. Pentagon officials say all three would accept an "all lawful purposes" standard. However, a senior administration official acknowledged competing models are "just behind" Claude Gov for specialized government applications.

[

SpaceX and xAI Enter Secret $100M Pentagon Contest for Autonomous Drone Swarms

SpaceX and its subsidiary xAI are competing in a secret Pentagon contest to build voice-controlled autonomous drone swarming technology, Bloomberg reported Sunday. The $100 million prize challenge, la

The Implicator

](https://www.implicator.ai/spacex-and-xai-enter-secret-100m-pentagon-contest-for-autonomous-drone-swarms/)

[

Europe's Trade Bazooka. Silicon Valley Holds Its Breath.

Thirty-six soldiers. That's what European nations sent to Greenland last week in a gesture of solidarity with Denmark. A training exercise, they called it. President Trump saw provocation. On Saturday

The Implicator

](https://www.implicator.ai/europes-trade-bazooka-silicon-valley-holds-its-breath/)

[

Europe Aims at Silicon Valley. The IMF Sees 2001.

San Francisco | January 20, 2026 Trump's Greenland tariffs just made the EU reach for its "economic nuclear weapon." The Anti-Coercion Instrument can revoke business licenses, ban government contract

The Implicator

](https://www.implicator.ai/europe-aims-at-silicon-valley-the-imf-sees-2001/)

• Vibe Coding Definition: This technique utilizes AI platforms, such as ChatGPT, to generate straightforward scripts that automate repetitive computer tasks without requiring prior programming knowledge.
• Automation Capabilities: Scripts created via vibe coding can perform actions such as combining files, extracting necessary information from large datasets, converting file types, and adjusting image dimensions.
• Getting Started: Beginners should initiate the process with small, manageable projects to ensure early success and build confidence.
• AI as a Coach: Users can instruct the AI platform to assume the role of a coding coach, guiding them step-by-step through the script creation and implementation process.
• Data Security Precaution: It is recommended to limit the AI's access to only a dedicated folder designated for vibe-coding projects to safeguard important data and system privacy.
• Simple Script Generation: A user can generate a script with a simple plain-language prompt, such as requesting a program to prepend a date to the beginning of every filename in a specified folder.
• Task Examples: Vibe coding is demonstrated as effective for consolidating PDFs into a single document, extracting email addresses from correspondence, and converting hyperlinks in a Word document to footnotes.
• Progression Path: After mastering simple scripts, users can gradually advance to more complex projects and potentially explore specialized tools like Claude Code for faster development and troubleshooting.

This copy is for your personal, non-commercial use only. Distribution and use of this material are governed by our Subscriber Agreement and by copyright law. For non-personal use or to order multiple copies, please contact Dow Jones Reprints at 1-800-843-0008 or visit www.djreprints.com.

https://www.wsj.com/tech/ai/vibe-coding-how-to-guide-211afcf1

Yes, You Can Vibe-Code. Here’s How to Get Started.

Creating your own programs might seem daunting. It’s a lot easier than you think.

By

Alexandra Samuel

Feb. 16, 2026 1:00 pm ET

Josie Norton for WSJ

• Vibe coding uses AI platforms like ChatGPT to generate simple scripts for automating tedious computer tasks without programming experience.

• Users can create scripts for tasks such as merging files, extracting information, converting file types and fixing image dimensions.

• To start, users should begin with small, simple projects. Ask the AI to act as a coding coach, and protect data by limiting AI access to a dedicated folder.

An artificial-intelligence tool created this summary, which was based on the text of the article and checked by an editor. Read more about how we use artificial intelligence in our journalism.

• Vibe coding uses AI platforms like ChatGPT to generate simple scripts for automating tedious computer tasks without programming experience.

  View more

Work is full of time-sucking, tedious or annoying tasks, particularly when you’re on a computer. I used to spend hours on stupid chores like reorganizing someone else’s messy spreadsheet.

Now, I use artificial intelligence to whip up programs that handle these tasks. What some people call “vibe coding” is a powerful way that AI can boost your efficiency and reduce work irritations.

You don’t need any programming experience because AI will do all the programming for you, and even coach you through the process. You can create a mini-program, known as a script, with a prompt as simple as, “Write a script that adds YYYY-MM-DD to the beginning of every file name in a folder, based on when it was created.” All you need to start vibe coding is a general-purpose AI platform like ChatGPT or Gemini. In a regular chat session, you tell the AI the problem you want to solve or the tool you need, and then it generates the code. (More on what you actually do with that code later.) Once you’ve created a script that solves your problem, you can use it anytime you need it.

You may have heard about tools designed for vibe coding, like Claude Code, which offer many more features and capabilities. But for most people, it’s wiser to start with simple scripts you copy, save and run yourself.

Vibe coding can sound intimidating, but it’s quickly becoming a common work tool for nonprogrammers—just like image creation and document layout used to be the exclusive domain of designers, but is now something many people do for themselves.

Here are some examples of some simple tasks you can do to start vibe coding—and then instructions on how to get going.

Sample tasks

• Merge files. It used to take me a few days to get all my different receipts into a single file for my accountant. Then I used vibe coding to write a script that takes all the PDFs in one folder and makes them into one big document. I have used similar scripts to combine many emails into one reference file, or to take a whole bunch of short documents and combine them into one big draft (with subheadings based on each file name).

• Extract information. Whenever I have to pull a small amount of information out of a huge file, I get AI to make a script. That’s how I took a folder full of past emails and extracted the email addresses of each person who emails me regularly. I also did tasks like rolling through a Word document full of hyperlinks, converting each of them to a footnote. Again, it’s as simple as typing: “Write a script that takes all the hyperlinks in this Word doc, and converts them to footnotes.”

• Clean and convert files. Do you have a bunch of PDFs that would be more useful as text files, or a text file that would be more useful to you as a spreadsheet? You can use vibe coding to write scripts that convert one file type to another.

• Fix finicky images. When I’m posting images to my blog or Instagram, they often need to be a specific size; I used AI to write a script that resized a whole folder full of images so that they would have consistent dimensions when I posted them on my website.

• Get rid of garbage. When I download a data file to my Mac, I often end up with a bunch of random garbage characters in the file; I used AI to write a little file-cleanup utility that fixes each downloaded file so that it opens nicely in Excel.

Tips to get started

Ready to get started? Here are a few strategies that will help:

• Start small. You don’t have to have a clear plan for how you want to try vibe coding. The next time you’re annoyed with a repetitive or boring task, ask an AI whether and how vibe coding could help. Be sure to ask whether this is a simple script or something that will likely require more patience and skill.

Start with simple projects that have a high likelihood of quick success, like asking your AI for a script that can combine files or perform a simple calculation. It should be something where you can describe your problem and the desired solution for the AI in a sentence or two, using plain language, like, “Write a script that will convert a folder full of Word docs to PDFs.”

The AI can walk you through what you need to do next, so you can actually get the script up and running.

• Set up a coding “coach.” Tell ChatGPT, Claude or another AI platform when and how you want it to help you write your own scripts. Explain your current level of tech knowledge: If you’ve never touched code or a script before, tell the AI you’re starting from square one and it should explain every single step you need to take. Whenever it gives you an instruction you don’t understand, say so; it will clarify.

• Increase gradually. As you get familiar with how scripts and programs are structured, you can take on more-complex projects. That is when you can level up by trying a platform like Claude Code, which makes it faster to write, troubleshoot and run your scripts.

• Protect your data, privacy and security. There is a risk that an AI could create and run a script that does something you don’t want—like harming files you care about, or compromising your privacy and security. You can ask the AI to flag any risks (and consider asking another AI for a second opinion), but it is always wise to back up your computer before you start a coding project, and to limit the AI’s access to a single folder that you use only for your vibe-coding projects.

Does this sound a bit daunting? If so, that’s yet another reason to give vibe coding a try. When you see how AI can help you learn and accomplish something that feels like it’s beyond your knowledge and skill set, that doesn’t just open the doors to writing your own programs: It opens the door to learning an ever-growing range of skills.

Write to reports@wsj.com

Workplace Technology

Read the full report

Workers Are Afraid AI Will Take Their Jobs. They’re Missing the Bigger Danger.

Technology Can Help People With Dementia Stay on the Job

People Trust Websites More When They Know a Human Is on Standby

How WSJ Readers Use AI at Work

Now That We Can Transcribe Work Meetings and Conversations, Should We?

How Workplace Tech Has Changed Over the Years, From the WSJ Archives

Doctors and Hospitals Look to Drones to Deliver Drugs, Supplies—and Even Organs

Office Technology Comes to the Movies: Test Your Knowledge With Our Quiz

Copyright ©2026 Dow Jones & Company, Inc. All Rights Reserved. 87990cbe856818d5eddac44c7b1cdeb8

By joining the conversation you are accepting our community rules and terms. To view rules, terms and FAQs**,** click here. Questions can be sent to moderator@wsj.com.

Conversations on news articles and news columns must be kept on the topic of the story. In opinion content, conversations can delve into other subjects. The conversation will close on this article four days after publication.

Be the first to comment...

No one seems to have shared their thoughts on this topic yetLeave a comment so your voice will be heard first.

[Powered by

](https://archive.is/o/WRYqG/https://www.openweb.com/powered-by)

||

Next in Journal Reports

[

Journal Reports

Workers Are Afraid AI Will Take Their Jobs. They’re Missing the Bigger Danger.

By Matthew Call

February 15, 2026 at 5:00 PM ET

It isn’t whether artificial intelligence is going to replace them. It’s who will control the knowledge that companies capture from their employees.

](https://archive.is/o/WRYqG/https://www.wsj.com/lifestyle/careers/ai-knowledge-capture-employees-a69a0e1c)

More Journal Reports Articles

[

Journal Reports

Technology Can Help People With Dementia Stay on the Job

By Heidi Mitchell

February 14, 2026 at 2:00 PM ET

](https://archive.is/o/WRYqG/https://www.wsj.com/lifestyle/workplace/aging-dementia-work-longer-ce417ffb)

[

Journal Reports

People Trust Websites More When They Know a Human Is on Standby

By Lisa Ward

February 13, 2026 at 7:00 PM ET

](https://archive.is/o/WRYqG/https://www.wsj.com/business/people-trust-humans-e37a6fa5)

[

Journal Reports

How Fixed-Rate Bonds Could Be Hurting Your Returns

By Derek Horstmeyer

February 13, 2026 at 6:50 PM ET

](https://archive.is/o/WRYqG/https://www.wsj.com/finance/investing/fixed-rate-bonds-investment-returns-2717a7d4)

Videos

Advertisement

Further Reading

[

U.S. Futures Fall, Global Markets Mixed; Bitcoin Recovers Slightly

](https://archive.is/o/WRYqG/https://www.wsj.com/finance/stocks/u-s-futures-fall-global-markets-mixed-bitcoin-recovers-slightly-d9cf31c2)

Advertisement

• Civic Ordinance Update: Barcelona has introduced a new municipal ordinance on coexistence that tightens sanctions and expands the definition of uncivil behavior to reinforce respect for public space.
• Increased Penalties: The updated regulations, replacing those from 2006, allow for fines up to €3,000 in the most serious instances.
• Mayor's Justification: Mayor Jaume Collboni stated the update aims to combat new problems and restore civic responsibility, asserting only those disrespecting the city need worry.
• Sexual Conduct Sanctions: New regulations introduce fines up to €600 for public indecency, public masturbation, or degrading verbal expressions.
• Alcohol-Related Offenses: Organizing or participating in "ethyl routes" can result in fines up to €3,000, while public alcohol consumption fines can reach €1,500 if minors are present or in restricted nighttime areas.
• Public Cleanliness Rules: Dog owners face fines up to €300 if they fail to dilute animal urine with water, and littering can incur sanctions up to €750.
• Severe Public Urination Penalties: Urinating or performing physiological functions in public streets, especially narrow ones, nightlife zones, or near public restrooms, can lead to fines up to €750.
• Payment Facilitation: The police force is equipped with dataphones for immediate sanction payment, and the city council is working to enable payment via Bizum, aiming to increase deterrence and collection efficiency.

Barcelona's new municipal ordinance on coexistence, known as the Civic Ordinance, introduces more prohibitions, toughens the sanctions regime, and expands the definition of uncivil behavior with the aim of reinforcing respect for public space. The regulations update those in force since 2006 and include fines that can reach €3,000 in the most serious cases.

Mayor Jaume Collboni has defended the update, stating that it aims to address new problems and restore civic responsibility. " Only those scoundrels who disrespect the city and break the rules" should be concerned about its enforcement, he affirmed this Monday at a press conference alongside the head of the Guardia Urbana (municipal police), Pedro Velázquez; the third deputy mayor for Security, Albert Batlle; and the commissioner for Coexistence, Montserrat Surroca.

According to the mayor, the ordinance constitutes “a major city agreement” that aims to update behaviors and increase penalties. “It's about restoring civic awareness, respect for the rules, and keeping in mind that living in a society and in a city offers many rights and advantages, but also obligations, commitments, and rules,” he emphasized.

The Barcelona City Police have already begun operations in key locations across all districts, particularly in nightlife areas, to enforce regulations on the ground. The city council stated that the goal is to "restore the excellence of public spaces" as part of its "360-degree security" strategy, which integrates cleanliness, community relations, and safety.

Sanctions

La nueva normativa incorpora medidas en ámbitos diversos como el consumo de alcohol, la protección de la dignidad sexual o las actividades incívicas organizadas. Algunas conductas que hasta ahora no estaban reguladas de manera específica pasan a ser sancionables.

Entre ellas, el exhibicionismo, la masturbación en público o las expresiones verbales degradantes podrán castigarse con multas de hasta 600 euros. También se prevén sanciones de hasta 300 euros por exhibir elementos sexuales en despedidas de soltero si se desoyen las advertencias policiales.

En el ámbito de las celebraciones, se consolida la prohibición de las llamadas “rutas etílicas”. Organizar o participar en estas actividades podrá suponer multas de hasta 3.000 euros. Además, el consumo de alcohol en la vía pública se endurece: los botellones, que ya se sancionaban con entre 100 y 600 euros, podrán alcanzar los 1.500 euros si se realizan en presencia de menores o en espacios con restricciones acústicas en horario nocturno.

La ordenanza también introduce sanciones para quienes intenten captar clientes para locales de consumo de productos cannábicos ilegales, una práctica que hasta ahora carecía de regulación específica.

Limpieza

Otra de las novedades afecta a la limpieza del espacio público. Los propietarios de perros podrán ser multados con hasta 300 euros si no diluyen con agua la orina de los animales. Asimismo, ensuciar la vía pública con vasos, envoltorios, chicles u otros residuos podrá comportar sanciones de hasta 750 euros.

Orinar o realizar necesidades fisiológicas en la calle también se penaliza con mayor severidad. En calles estrechas, zonas de ocio nocturno o áreas con lavabos públicos disponibles, la multa podrá alcanzar los 750 euros.

La normativa endurece igualmente las sanciones contra las pintadas y grafitis. No solo se castigará su realización en espacios no autorizados, sino también la promoción de estas prácticas, con multas de 600 euros, además de la obligación de asumir los costes de limpieza y reparación de los daños.

La ordenanza, publicada ya en el Boletín de la Provincia de Barcelona, introduce cambios en los plazos de pago. Si la sanción se abona en los dos primeros días hábiles tras su notificación, se aplicará una reducción máxima del 50%. En un segundo plazo, la rebaja será del 40%.

En paralelo, el Ayuntamiento mantiene el pago en cajeros de entidades bancarias, pero refuerza el cobro inmediato. Los agentes disponen de datáfonos para abonar las sanciones en el momento y el consistorio trabaja para habilitar el pago por Bizum. Según Velázquez, este sistema busca aumentar el carácter disuasorio y facilitar el cobro, especialmente en el caso de personas no residentes.

Balance

The head of the Barcelona Municipal Police explained that 114,528 complaints related to antisocial behavior were registered in 2025, meaning that "one in four incidents" handled by the 112 emergency services were related to such conduct. In total, the force responded to 422,798 incidents, 5.2% more than the previous year, of which 34,089 involved neighborhood disputes.

The municipal information campaign about the ordinance, which began weeks ago, will continue indefinitely with the aim of disseminating the new developments and promoting compliance.

• Strategic Acquisition: OpenAI secured Peter Steinberger, creator of the OpenClaw framework, following a competitive bidding process involving Meta and multi-billion dollar valuation offers.
• Market Traction: The acquisition was driven by OpenClaw's rapid adoption, including 196,000 GitHub stars and 2 million weekly visitors, rather than the technical quality of its 300,000-line codebase.
• Security Vulnerabilities: OpenClaw has been linked to significant security failures, including the leak of 1.5 million API keys and the distribution of malware via its skills marketplace, ClawHub.
• Economic Independence: Steinberger, having previously sold a company for over $100 million, operated OpenClaw at a personal loss of $10,000 to $20,000 monthly before joining OpenAI.
• Industry Disruption: The popularity of local AI agents through OpenClaw significantly impacted hardware markets, causing six-week delays for high-memory Apple Mac configurations.
• Philosophical Divergence: Despite joining OpenAI, Steinberger publicly rejects the pursuit of Artificial General Intelligence (AGI) in favor of specialized, collaborative agent systems.
• Pragmatic Distribution: Steinberger prioritized OpenAI's massive distribution network and 300 million weekly users over Meta's personal engagement and direct feedback from Mark Zuckerberg.
• Competitive Shift: Although Anthropic’s models originally powered OpenClaw, the project’s migration to OpenAI aligns the most prominent agent brand with a primary competitor.

Mark Zuckerberg needed ten minutes. He was finishing code.

Peter Steinberger had called him on WhatsApp without scheduling anything. "I don't like calendar entries," he told Lex Fridman last week. "Let's just call now." Zuckerberg asked for a brief pause, then picked up. The first ten minutes devolved into an argument about whether Claude Code or Codex was the better programming tool. The CEO of a trillion-dollar company squabbling with a solo developer from Vienna over IDE preferences.

That was two weeks ago. Zuckerberg ran OpenClaw on his own machine afterward. Gave feedback that was blunt and specific, calling features "great" or "shit" in real time. Used it until it broke, then sent notes on what to fix. Steinberger called it "the biggest compliment" because "it shows they actually care about it."

On Sunday, Steinberger announced he's joining OpenAI. Not Meta. Sam Altman posted within hours that Steinberger would "drive the next generation of personal agents." Greg Brockman and Fidji Simo both posted within the hour. Three executives in a coordinated burst of enthusiasm that OpenAI hasn't shown for a hire since the $6.4 billion Jony Ive acquisition last year.

Every outlet is covering this as a talent acquisition win. The framing is wrong. OpenAI already employs thousands of good engineers. What it couldn't build internally is proof that ordinary people will hand an AI agent full access to their digital lives without hesitation.

What $10,000 a month buys at the negotiating table

The Breakdown

• OpenAI hired OpenClaw creator Peter Steinberger; both Meta and OpenAI reportedly made offers valued in the billions

• OpenClaw's 300,000+ line codebase was never the prize; 196,000 GitHub stars and 2 million weekly visitors were

• Steinberger publicly rejects AGI but chose OpenAI's distribution over Meta's personal attention from Zuckerberg

• Anthropic's models powered OpenClaw, but the framework's creator now works for the competition

Steinberger has been losing money on OpenClaw since November. Ten to twenty thousand dollars a month, by his own count. He routes sponsorship revenue to the developers who maintain his dependencies rather than keeping it. The project hit 196,000 GitHub stars and pulled 2 million visitors in a single week while its creator subsidized everything from savings.

He didn't need their money. He spent 13 years building PSPDFKit into a PDF tools company worth over $100 million before selling to Insight Partners. Three years of what he described as soul-searching followed. Therapy. Ayahuasca. Forty-three failed projects. Then OpenClaw caught fire.

His negotiating edge came from something no check could replicate. Both Meta and OpenAI made concrete offers, reportedly valued in the billions. VCs lined up. Steinberger told Fridman he doesn't care. "I don't give a fuck" were his exact words. When you've already sold a company and your next project goes viral by accident, the dynamics flip completely. He wasn't selling. They were auditioning.

The code nobody wanted

If you think these companies wanted OpenClaw's codebase, look at the inventory. Somewhere north of 300,000 lines of code, nobody is sure exactly how many. Unaudited. Developer Gavriel Cohen evaluated it for NanoClaw, found it too bloated for any security team to review properly, and rebuilt the core logic in roughly 500 lines of TypeScript. His team audited the entire replacement system in an afternoon.

The security record is just as ugly. Moltbook leaked 1.5 million API keys to the open internet. ClawHub, OpenClaw's skills marketplace, hosted 335 packages distributing Atomic Stealer malware to Mac users. Three days after Moltbook launched, RentAHuman.ai went live, a marketplace where OpenClaw agents hire real humans for physical tasks. Forty thousand people signed up to take orders from bots. Payment in stablecoins.

No serious engineer looks at that inventory and says ship it. What Meta and OpenAI wanted is the adoption graph. Six hundred contributors. Ten thousand commits. All in under three months. The WhatsApp integration dropped an AI agent into a messaging app 3 billion people carry in their pockets. OpenClaw became the first consumer brand in AI agents without spending a dollar on marketing.

OpenClaw's momentum even bent hardware markets. Tom's Hardware reported that Mac delivery times for high-memory configurations stretched to six weeks, driven partly by users buying machines to run local AI agents. Apple CEO Tim Cook acknowledged the company was chasing memory supply. One solo developer's weekend project created enough demand to disrupt Apple's supply chain.

You can rewrite 400,000 lines of messy code in a quarter. You cannot fabricate that kind of pull. That distinction, according to the bidding behavior of two of the most powerful companies in technology, is worth billions.

Specialized intelligence inside the AGI machine

The deal gets uncomfortable when you look at what Steinberger actually believes.

He told a Y Combinator podcast this month that AGI is the wrong goal. "What can one human being actually achieve? Do you think one human being could make an iPhone or one human being could go to space?" he said. "As a group we specialize, as a larger society we specialize even more." His vision runs on specialized agents collaborating. Not one god model that handles everything.

OpenAI's entire corporate identity rests on achieving artificial general intelligence. The name says it. So does the $500 billion valuation. Steinberger just joined a company whose stated mission he publicly rejects.

Ignore the ideology for a second. The logistics explain everything. Steinberger does not want to run a company. Thirteen years of it was enough. He wants to build agents everyone can use. That means compute, APIs, and 300 million people already opening ChatGPT every week. This is pragmatism dressed as alignment.

Meta offered something different. More personal, in fact. Zuckerberg's hands-on engagement impressed Steinberger. Steinberger recounted it on Fridman's show, acting out the reactions. "Mark basically, 'Oh, this is great. Oh, this is shit. Oh, it needs to change this.'" He noted the contrast with OpenAI. "I didn't get the same on the OpenAI side."

Stay ahead of the curve

Strategic AI news from San Francisco. No hype, no "AI will change everything" throat clearing. Just what moved, who won, and why it matters. Daily at 6am PST.

No spam. Unsubscribe anytime.

He chose OpenAI anyway. Zuckerberg codes, gives real feedback, and clearly cares. None of that ships a product to hundreds of millions of users. For a builder who wants reach without management overhead, that settled it.

Chrome won. Chromium exists.

Steinberger drew the comparison himself. OpenClaw would follow the Chrome and Chromium model. A foundation to hold the open-source project. A corporate partner to build the commercial version. "I think this is too important to just give to a company and make it theirs," he told Fridman.

He's describing a pattern with a known ending.

Google open-sourced Chromium. Chrome is what everyone downloads. Android AOSP is open. Google's Android with Play Services runs the planet. MySQL went to Oracle. The community forked it into MariaDB. MariaDB survives. MySQL still owns the market.

Every foundation arrangement in tech follows the same gravitational pull. The corporate version gets full-time engineers, marketing, distribution, and the daily attention of the person who created the project. The open-source twin gets volunteers and good intentions. Gravity always wins.

Altman committed in public. "OpenClaw will live in a foundation as an open source project that OpenAI will continue to support." Those words are sincere until they collide with product timelines. When OpenAI ships a consumer agent built on Steinberger's ideas, using OpenAI's models, and that product competes with what the foundation maintains, the foundation will not get the best features first. OpenAI has shareholders. Shareholders beat charters every time.

Steinberger is too experienced to miss this. Thirteen years running a company taught him what public commitments look like right before they erode. The foundation language gives both sides something comfortable to say in public. It hands the community a story to hold onto. And it gives OpenAI a grace period before anyone asks which version is getting Steinberger's best hours.

Anthropic built the runway. OpenAI caught the flight.

Nobody is talking about who this hurts most.

OpenClaw launched on Claude. Steinberger called himself "the biggest unpaid Codex advertisement show," but the framework's recommended setup pointed users to "Anthropic Pro/Max plus Opus 4.5 for long-context strength and better prompt-injection resistance." Thousands of OpenClaw users routed heavy agent workloads through Anthropic's $20 consumer subscriptions.

Then Anthropic pushed back. Steinberger said in a recent interview that Anthropic "doesn't like it anymore." He recommended API keys instead. The economics made sense from Anthropic's side. A subscription designed for individual chat sessions was never priced for autonomous agents burning through tokens at 3 AM while their owner slept in the next room.

But the result is a category disaster. Anthropic's models powered the most visible agent framework in history. Its subscriptions funded the consumer proof of concept. And now the creator of that framework works for a competitor. He announced the move on a weekend while Anthropic's $380 billion valuation round closed the same week.

OpenClaw users will keep running Claude. The models are good. But the face of the agent movement now works at a rival lab, and every keynote, podcast, and product launch will carry OpenAI's logo behind him.

Anthropic's discomfort won't be loud. It will be the quiet kind. The kind where you helped prove that agent interfaces matter more than models, then watched someone gift-wrap that proof for the competition. At $380 billion, Anthropic can absorb the sting. What it cannot do is un-train a community that learned to think of AI agents through OpenClaw's interface, an interface now housed inside OpenAI's strategy.

The gravity starts now

Zuckerberg is still coding at Meta. He'll build agents regardless. Google has the largest mobile distribution surface on earth through Android and has done nothing visible with consumer agents. Apple has iMessage, Siri, and 1.5 billion active devices. Silence.

The foundation will publish its charter. Volunteers will submit pull requests. Steinberger will show up at community events and champion the open-source spirit he clearly values. He means it. That sincerity is real.

And slowly, because this is how gravity works in tech, the best ideas and the best engineering hours will flow toward OpenAI's agent products. Steinberger spends his days there now. So will his attention. So will the features that matter.

Chrome won. Chromium exists. The claw is the law, until corporate gravity says otherwise.

Frequently Asked Questions

Q: Why did Steinberger choose OpenAI over Meta?

A: Despite Zuckerberg's hands-on engagement, Steinberger chose OpenAI for its distribution infrastructure. OpenAI has 300 million weekly ChatGPT users and can ship agent products at scale. Steinberger wanted reach without management overhead.

Q: What happens to OpenClaw as an open-source project?

A: Steinberger and Altman committed to a foundation model where OpenClaw remains open-source while OpenAI builds a commercial version. The pattern mirrors Chrome and Chromium, where the corporate version historically absorbs the best features and talent.

Q: How much was the Steinberger acquisition reportedly worth?

A: Both Meta and OpenAI made offers reportedly valued in the billions. VCs also lined up with proposals. Steinberger, who previously sold PSPDFKit for over $100 million, said he did not care about the money.

Q: What security issues has OpenClaw faced?

A: Moltbook leaked 1.5 million API keys. ClawHub hosted 335 packages distributing Atomic Stealer malware to Mac users. RentAHuman.ai launched days later, letting AI agents hire humans for physical tasks with stablecoin payment.

Q: How does the Steinberger hire affect Anthropic?

A: OpenClaw launched on Claude and recommended Anthropic subscriptions for agent workloads. Thousands of users routed heavy tasks through Anthropic's $20 plans. The framework's creator now works at a rival lab during Anthropic's $380 billion valuation round.

[

Former GitHub CEO Thomas Dohmke Raises Record $60M for New Startup Entire

Thomas Dohmke, who stepped down as CEO of Microsoft's GitHub in August 2025, came out of stealth on Tuesday with Entire, a new developer platform built for the age of AI coding agents. The startup rai

The Implicator

](https://www.implicator.ai/former-github-ceo-thomas-dohmke-raises-record-60m-for-new-startup-entire-2/)

[

OpenAI Launches Codex Desktop App for macOS With Multi-Agent Workflows and Doubled Rate Limits

OpenAI released a macOS desktop app for Codex today, turning its AI coding agent into a standalone application that can run multiple agents across different projects at the same time. The company also

The Implicator

](https://www.implicator.ai/openai-launches-codex-desktop-app-for-macos-with-multi-agent-workflows-and-doubled-rate-limits/)

[

OpenAI Launches Frontier to Manage AI Agents From Rival Vendors in One System

OpenAI on Thursday released Frontier, a platform that lets companies build, deploy, and manage AI agents from multiple vendors inside a single system. The product works with agents built by OpenAI, by

The Implicator

](https://www.implicator.ai/openai-launches-frontier-to-manage-ai-agents-from-rival-vendors-in-one-system/)

New Hampshire’s state lines are dotted with shopping malls. The Pheasant Lane Mall’s parking lot is largely located in Massachusetts, though the mall itself sits within the Live Free or Die State. Stores cluster on the east side of the Connecticut River in New Hampshire, though the main interstate, I-91, runs along the west side of the river in Vermont. To shoppers, the reason is obvious: New Hampshire has no sales tax. As the owner of the state-line-adjacent Mall at Rockingham Park notes, you can “Shop TAX FREE all year long” at the stores “conveniently located just over the Massachusetts border.” The Pheasant Lane Mall even removed a cornerstone that would have extended a few feet over the border, avoiding contact with the state once known as “Taxachusetts.”

In recent years, consumers have had even more incentive to cross state lines in search of lower taxes. During the Great Recession in 2009, Massachusetts raised its sales tax from 5 percent to 6.25 percent; Maine followed in 2013, increasing its rate from 5 percent to 5.5 percent. Post-Covid inflation has driven up the price of goods—along with the amount of sales tax owed—even as incomes have lagged.

Since the pandemic, the cost of living has become the defining issue in American politics. Yet as inflation surged, few politicians targeted one of the most direct and controllable costs they impose on everyday purchases: the sales tax. Unlike property and income taxes, which have periodically provoked revolts, sales taxes have rarely faced organized political opposition. That helps explain why they are the one major tax category whose rates have risen almost continuously over the past century. Taken together, general sales taxes and selective sales taxes—special levies on goods such as cigarettes or rental cars—now constitute the largest source of revenue raised by state and local governments. Politicians truly concerned about the cost of living could start by reducing the one charge that most directly increases it.

States began imposing general sales taxes during the Great Depression. Cratering property-tax revenues led Mississippi to levy the first one in 1930. By 1950, 28 states had them, mostly taxing sales at about 2 percent. In the coming decades, all except five states would impose them (Vermont was the last to adopt one, in 1969), and the rate kept ratcheting upward. California’s sales-tax path is instructive: the tax started during the Depression at 2.5 percent, hit 4 percent in the 1960s, and had climbed to over 6 percent by the early 2000s. When combined with a mandatory sales tax collected by local governments, the rate is 7.25 percent.

Since getting authority from the states to enact their own sales taxes, localities’ rates have followed a similar upward path. In 1935, New York became the first city to authorize a general sales tax. Its one-cent, or 1 percent, rate had jumped to 3 percent by the early 1950s and now stands at 4.5 percent, plus a small extra sales tax for transit. When combined with the state rates, the city takes nearly 9 percent from shoppers. Thirty-eight states now allow local governments to impose their own sales taxes. In California, cities and counties can levy local sales taxes on top of state-mandated ones, which can push the combined state and local rate as high as 9.25 percent.

Sales-tax revenues exploded after the Supreme Court’s 2018 South Dakota v. Wayfair decision, which allowed state and local governments to charge sales taxes for online purchases. Internet retailers now must contend with more than 12,000 separate state and local sales-tax jurisdictions. The decision also spawned a host of new companies that help businesses navigate the tax maze, for fees that can range up to hundreds of thousands of dollars. (See “The Tax Nexus Cometh,” Spring 2023.) All states with sales taxes have expanded them to include online or “remote” sales, bringing in tens of billions in extra revenue by 2021.

In that year, states and local governments collected nearly $700 billion in sales taxes. Most of these were general sales taxes, covering all types of products. But just over $200 billion flowed from “selective” sales taxes, especially on alcohol, cigarettes, and gasoline. Coinciding with the inflation spike in 2021, sales taxes began climbing even more rapidly, with state sales-tax revenues surging 10 percent. In 2022, the increase was 14 percent. By 2023, even as inflation eroded the real value of state corporate and personal income-tax receipts, real sales-tax revenues kept climbing. Inflation made sales taxes a highly effective revenue tool—but it also made consumers more determined to avoid them.

As many Massachusetts residents will remind you, anyone buying goods in New Hampshire is technically required to pay a “use” tax to their home state. Almost no one does. Years ago, Massachusetts sued Town Fair Tire, a New Hampshire retailer just across the border, in an effort to obtain records on its clearly out-of-state customers. In response, New Hampshire passed a law, sponsored by then-state and now-U.S. senator Maggie Hassan, making it illegal for stores to share customer information with other states’ tax authorities. Town Fair Tire and its customers remained inviolate.

Today, only Delaware, Alaska, Montana, Oregon, and New Hampshire lack a general sales tax—and they’re not shy about advertising it. Drivers into Delaware were once greeted with the sign “Home of Tax-Free Shopping,” printed in bigger and bolder letters than Delaware’s previous claim to fame of being “The First State.” Following Wayfair, many people posted online threads asking how to get a shipping address in one of the tax-free states. One company offers a service for businesses to route products through Oregon or Delaware to avoid intermediate sales taxes—those charged when a firm buys goods before using them in manufacturing or resale. The company Global Shopaholics provides customers with a Delaware shipping address, allowing buyers from other countries to send their purchases there first before the goods get forwarded abroad tax-free. Other states’ tax authorities lament such arrangements, but they mostly reflect the widening gap between taxed and tax-free states.

Though economists generally tout the sales tax as an efficient way to raise revenue, with fewer distortions and loopholes than income taxes, the tax has become more riddled with exceptions and special rates over time. Currently, about 60 percent of all sales go untaxed, meaning that the remaining goods must bear a much higher rate. The reason: sales taxes mainly apply to physical goods, such as cars or electronics, but generally ignore services, such as haircuts or dental care, which constitute a growing share of the U.S. economy.

States often offer one-off exemptions to benefit certain groups. Many states provide exemptions for the necessities of clothing, food, and prescription drugs, for example, but others give a pass to flags, newspapers, feminine hygiene products, and renewable energy products. Deciding whether a good falls into a state’s exemption can require firms to exercise Talmudic intricacy. Wisconsin once issued guidance explaining which types of ice cream cake were taxable. The inclusion of utensils, or even a layer of fudge, could transform the dessert from a nontaxable food item into a taxable indulgence.

States have also used sales-tax exemptions to favor specific areas. Rather than competing with neighboring Delaware by cutting its general tax rate, New Jersey in 1983 created special Urban Enterprise Zones, allowing businesses in designated “underprivileged” areas to collect only half the state sales tax. In practice, the policy mainly benefits large retailers that draw customers from elsewhere in the state. Trenton’s enterprise zone became the surprising home to one of the largest Steinway piano dealers in the U.S., whose chief estimated that 80 percent of customers came from out of town.

For years, economists and policy experts, such as those at the Tax Foundation, have urged governments to “broaden the base and lower the rate,” meaning that they should tax more kinds of sales, especially services, while reducing overall rates. A rare success came in Washington, D.C., which in 2014 expanded its sales tax to cover services such as yoga studios and gyms and used the new revenue to cut income taxes and other levies. This proposal garnered support from an unusually broad coalition, ranging from the left-leaning Citizens for Tax Justice to Grover Norquist’s conservative Americans for Tax Reform.

More often, states have expanded the range of taxable services without lowering rates. Though the Wayfair ruling primarily addressed whether online retailers must collect sales tax, it also cleared the way for taxing all online transactions. Since then, many states have enacted taxes on digital downloads, streaming services, software subscriptions, and video games. States and cities have broadened selective sales taxes—imposing higher, separate rates on prepared meals, vending-machine sales, hotel stays, rental cars, cell phones, and live entertainment.

The purpose of the sales tax is to raise revenue from personal consumption—people spending money for their own enjoyment. Yet transactions between businesses often get taxed as well. Though states have tried to limit intermediate taxes—companies paying taxes on sales to each other—one state-commissioned estimate found that over 40 percent of total sales taxes came from business-to-business sales. Beyond distorting business decisions—since firms pay tax when they buy a product but not when they produce it themselves—these taxes can “pyramid,” with the same item taxed multiple times at different stages of production. The added costs are ultimately wrapped into the final price, even if the shopper never sees them on the receipt.

Progressives have long railed against sales taxes as regressive, disproportionately burdening the poor; but in recent years, they’ve readily supported higher local sales taxes—so long as the revenue funds their political priorities.

Sales taxes have become a popular way to pay for homeless services and subsidized housing, for instance. In 2024, Los Angeles County approved a half-cent sales tax for homeless housing and services, which was expected to generate over $1 billion annually—with no sunset date, as is typical for local tax measures. The fact that L.A. had already enacted a quarter-cent sales tax for the same purpose just seven years earlier—and that it produced no visible improvement—did little to dissuade local politicians or the county’s notably progressive voters. Politicians and voters ignored how previous sales-tax revenues were spent on apartments that averaged $600,000 per unit and whose construction was rife with corruption, as shown by the indictment of a city councilman who accepted bribes from prospective developers of homeless housing. Denver adopted a sales tax for homeless initiatives in 2020; the city failed to pass another such measure last year only because voters instead approved a sales tax to subsidize health care.

Many jurisdictions now ask voters to approve separate sales taxes to fund transit. Just before Los Angeles passed its first sales tax for homelessness, the county enacted a half-cent sales tax for transportation and transit. Since then, transit use has fallen by about one-fourth. In 2020, Seattle likewise raised its sales tax to support transit projects. That didn’t stop transit ridership from dropping by one-fourth from pre-pandemic levels.

After the pandemic period’s steep drop in ridership, many transit agencies, heedless of the strain on inflation-burdened consumers, sought more revenue rather than cut services to reflect diminished demand. In 2024, Columbus, Ohio, and Nashville, Tennessee, authorized half-cent tax increases to fund their transit systems. Mecklenburg County, the home of Charlotte, North Carolina, passed a one-cent tax hike for transit in November 2025.

While voters must approve most of these local sales taxes, government agencies try to obfuscate where the money is going. Los Angeles said that the first goal of its sales tax for transportation was to “improve freeway traffic flow” and that another objective was to “repave local streets, repair potholes, synchronize signals.” But buried deep in the spending plan, the government acknowledged that only 17 percent of the funds were going to roads; the rest went to transit and more niche travel modes like bicycle paths. Other governments have tried to remove voters entirely from tax decisions. In 2020, Washington State gave local governments the power to impose a sales tax for affordable housing without submitting the proposal for voter approval.

The enduring mystery of the sales tax is why it never seems to go down. Other levies face frequent taxpayer revolts, but the hit to consumers from a penny sales tax is apparently abstract enough that most don’t notice it. Louisiana made one of the rare sales-tax reductions in recent years, in 2018, reducing its top rate by over half a cent. But this year, it returned to its previous rate of 5 cents as part of a general tax reform.

Even when politicians talk about the cost of living, the sales tax rarely comes up. Zohran Mamdani won the mayoral election in New York largely by promising to bring living costs under control, and other progressive city politicians have followed his lead. Yet none has suggested cutting the 8.875 percent surcharge that government adds to purchases. Instead, progressives in New York, like their counterparts nationwide, have pushed for new consumer taxes to fund their priorities, even while touting their affordability agendas.

The steady rise of sales taxes, along with their growing complexity, adds to the burden on businesses and consumers already strained by inflation. Politicians could act to ease that burden. It remains striking how few seem interested in doing so.

This article is part of “An Affordability Agenda,” a symposium that appears in City Journal’s Winter 2026 issue.

Photo: Few politicians have targeted levies on consumer purchases as a way of reducing prices. (Lindsey Nicholson/UCG/Universal Images Group/Getty Images)

• Software Pricing Shift: A prediction suggests software billing may move toward a pay-per-outcome model, where payment is contingent on an AI agent achieving a specific objective.
• AI Insider Concerns: Individuals inside AI companies are issuing numerous warnings regarding the accelerating sophistication and potential real-world harms of artificial intelligence.
• Specific AI Dangers: Warnings mentioned include autonomous cyberattacks, mass unemployment due to disruption, and the replacement of human relationships by AI.
• Researcher Departures: An Anthropic researcher resigned, citing that the “world is in peril” from AI, among other threats.
• OpenAI Staff Discontent: Some OpenAI staffers expressed concerns over plans to introduce erotica and the potential for manipulation arising from ad integration.
• Accelerated Advancement: The urgency surrounding AI warnings is attributed to the rapid advancement of AI capabilities surpassing the expectations of seasoned researchers.
• Existential Threat Concern: An OpenAI staffer voiced feeling an "existential threat" from AI, questioning what work will remain for humans when AI becomes highly proficient.
• Columnist Topics: Featured columns cover privacy issues with home security cameras, investment focus on AI infrastructure providers, and the billionaire competition in space exploration.

By

Georgia Wells

Feb. 15, 2026 10:59 am ET

Your browser does not support HTML5 video.

Why Software Pricing May Move to Pay-Per-OutcomePlay video: Why Software Pricing May Move to Pay-Per-Outcome

Keep hovering to play

Sierra Co-founder and CEO Bret Taylor discusses the future of software billing, predicting a move away from monthly licenses to a model where companies only pay when an AI agent successfully completes a job or closes a sale. Photo: WSJ Leadership Institute

This is an edition of the WSJ Technology newsletter, a weekly digest of tech columns, big stories and personal tech advice. If you’re not subscribed, sign up here.

Tech insiders are sounding an alarm.

The accelerating sophistication of artificial intelligence is driving a wave of warnings that AI can create real-world harms, including autonomous cyberattacks, mass unemployment, unrelenting market disruption and the replacement of human relationships.

Newsletter Sign-up

Technology

A weekly digest of tech columns, big stories and personal tech advice, plus a news ticker and a touch of dark humor.

Subscribe

A researcher at Anthropic this week said he is leaving the company, writing in a letter to colleagues that the “world is in peril” from AI, among other dangers. Inside OpenAI, some staffers have voiced concerns about the company’s plan to roll out erotica. Another OpenAI researcher said she was quitting OpenAI, citing its plan to introduce ads and her fear that the company would face huge incentives to manipulate users.

Artificial intelligence joins a long list of industries that have prompted dire insider warnings. But the sirens about AI are occurring earlier in the industry’s development, and in a greater volume, relative to other technological revolutions.

Some of the urgency can be traced to the rapid advancement of AI capabilities, which has surprised some of the most seasoned researchers and coders.

“Today I finally feel the existential threat that AI is posing,” OpenAI staffer Hieu Pham wrote on X Wednesday. “When AI becomes overly good and disrupts everything, what will be left for humans to do?”

—Georgia is a tech reporter based in San Francisco.

The Latest From Our Columnists

[

](https://www.wsj.com/tech/personal-tech/ring-nest-home-security-cameras-privacy-157935f8)

ELENA SCOTTI/WSJ; ISTOCK

Nicole Nguyen: The Dragnet Era of Home Security Cameras

Most people get security cameras for, well, security. They hope the devices will deter criminals, or at least catch bad guys in the act. But fundamental privacy questions, like what happens to the videos and who owns the footage, should be part of the calculus when installing the devices. Recent events prove the point.

Dan Gallagher: Picks and Shovels Still Rule the AI Tech Trade

The AI trade has definitely become more fraught. But at least one constant remains: Investors are choosing the companies on the receiving end of big tech’s spending spree.

Tim Higgins: Bezos vs. Musk: The New Billionaire Battle for the Moon

The contest between Elon Musk and Jeff Bezos is only going to get more heated now that the two are directly competing for the moon. The faceoff promises to stoke an even hotter 21st-century space race—this time between this era’s real superpowers: billionaires.

Big Stories

[

](https://www.wsj.com/tech/ai/anthropic-amanda-askell-philosopher-ai-3c031883)

Lindsay Ellary for WSJ Magazine

Meet the One Woman Anthropic Trusts to Teach AI Morals

Amanda Askell knew from the age of 14 that she wanted to teach philosophy. What she didn’t know then was that her only pupil would be an AI chatbot named Claude.

As the resident philosopher of the tech company Anthropic, Askell spends her days learning Claude’s reasoning patterns and talking to the AI model, building its personality and addressing its misfires with prompts that can run longer than 100 pages.

Inside OpenAI’s Decision to Kill the AI Model That People Loved Too Much

ChatGPT’s 4o model was beloved by many users, but controversial for its sycophancy and real-world harms linked to some conversations

🖨️ Tech Ticker

[

](https://www.wsj.com/tech/ai/the-ai-gold-rush-is-breaking-a-silicon-valley-taboo-cashing-out-before-the-ipo-4844f6c1)

Emil Lendof/WSJ, iStock

Cashing Out Is In: The AI gold rush is breaking a Silicon Valley taboo: cashing out before the IPO, an option Stripe, OpenAI, Anthropic, Databricks and SpaceX are giving employees.

New Anthropic Director: Former Microsoft and General Motors executive Chris Liddell, who helped take the automaker public, has joined Anthropic’s board of directors.

Musk’s xAI Reorg: Elon Musk announced a reorganization of xAI, his artificial-intelligence startup, just days after merging it with SpaceX, his rockets-and-satellites business.

Other Smart Stuff

• Meta Plans to Add Facial Recognition Technology to Its Smart Glasses (The New York Times)
• I Tried RentAHuman, Where AI Agents Hired Me to Hype Their AI Startups (Wired)
• Waymo Is Getting DoorDashers to Close Doors on Self Driving Cars (404 Media)

🎥 Watch This: Bret Taylor Discusses the Future of the Software Business

[

](https://www.wsj.com/tech/ai/when-ai-bots-start-bullying-humans-even-silicon-valley-gets-rattled-0adb04f1)

Dado Ruvic/Reuters

Tech News Briefing

We’re Doomed…

[

](https://www.wsj.com/lifestyle/relationships/stanford-students-experiment-dating-date-drop-92a4aea8)

Screenshot showing Date Drop questionnaire

When Stanford students go wild for an algorithm that can find the perfect match.

About Us

The Technology newsletter is a weekly digest of tech reviews, columns and headlines from Deputy Tech & Media Editor Wilson Rothman and Deputy Tech Bureau Chief Brad Olson. Write to Wilson at wilson.rothman@wsj.com and Brad at bradley.olson@wsj.com. Got a tip for us? Here’s how to submit.

Copyright ©2026 Dow Jones & Company, Inc. All Rights Reserved. 87990cbe856818d5eddac44c7b1cdeb8

What to Read Next

[

Meet Anthropic’s AI Morality Teacher

](https://www.wsj.com/tech/ai/meet-anthropics-ai-morality-teacher-b8010921?mod=WTRN_pos1)

[

Plus, home security cameras enter their dragnet era and a Stanford student’s matchmaking algorithm takes over campus.

](https://www.wsj.com/tech/ai/meet-anthropics-ai-morality-teacher-b8010921?mod=WTRN_pos1)

Continue To Article

[

When AI Bots Start Bullying Humans, Even Silicon Valley Gets Rattled

](https://www.wsj.com/tech/ai/when-ai-bots-start-bullying-humans-even-silicon-valley-gets-rattled-0adb04f1?mod=WTRN_pos2)

[

An extraordinary example of online aggression by a bot is contributing to fears of real-world harm caused by artificial intelligence.

](https://www.wsj.com/tech/ai/when-ai-bots-start-bullying-humans-even-silicon-valley-gets-rattled-0adb04f1?mod=WTRN_pos2)

Continue To Article

EXCLUSIVE

[

Anthropic Adds New Board Member as It Eyes IPO

](https://www.wsj.com/tech/ai/anthropic-ai-board-chris-liddell-1df5545b?mod=WTRN_pos4)

[

The former Microsoft and GM executive Chris Liddell has previously worked for the Trump administration.

](https://www.wsj.com/tech/ai/anthropic-ai-board-chris-liddell-1df5545b?mod=WTRN_pos4)

Continue To Article

[

Musk Announces xAI Reorganization, Staff Departures

](https://www.wsj.com/tech/musk-announces-xai-reorganization-staff-departures-d3e0fbf9?mod=WTRN_pos5)

[

The co-founders exit as the artificial intelligence startup merges with SpaceX.

](https://www.wsj.com/tech/musk-announces-xai-reorganization-staff-departures-d3e0fbf9?mod=WTRN_pos5)

Continue To Article

[

A Stanford Experiment to Pair 5,000 Singles Has Taken Over Campus

](https://www.wsj.com/lifestyle/relationships/stanford-students-experiment-dating-date-drop-92a4aea8?mod=WTRN_pos6)

[

A student built a matchmaking algorithm that has consumed the school—and highlighted the challenges of finding love for high achievers.

](https://www.wsj.com/lifestyle/relationships/stanford-students-experiment-dating-date-drop-92a4aea8?mod=WTRN_pos6)

Continue To Article

[

Inside OpenAI’s Decision to Kill the AI Model That People Loved Too Much

](https://www.wsj.com/tech/ai/chatgpt-4o-openai-315138b8?mod=WTRN_pos7)

[

ChatGPT’s 4o model was beloved by many users, but it was controversial for its sycophancy and the real-world harms linked to some conversations.

](https://www.wsj.com/tech/ai/chatgpt-4o-openai-315138b8?mod=WTRN_pos7)

Continue To Article

[

Senior AI staffers keep quitting — and are issuing warnings about what’s going on at their companies

](https://www.marketwatch.com/story/senior-ai-staffers-keep-quitting-and-are-issuing-warnings-about-whats-going-on-at-their-companies-f770f785?mod=WTRN_pos8)

[

Through social-media posts and even a resignation letter in the New York Times, former employees of companies like OpenAI and Anthropic are not leaving quietly.

](https://www.marketwatch.com/story/senior-ai-staffers-keep-quitting-and-are-issuing-warnings-about-whats-going-on-at-their-companies-f770f785?mod=WTRN_pos8)

Continue To Article

[

350-Year-Old Coral-Colored House That’s Hosted Prince Harry and Helen Mirren Lists for Sale in Barbados

](https://www.mansionglobal.com/articles/350-year-old-coral-colored-house-thats-hosted-prince-harry-and-helen-mirren-lists-for-sale-in-barbados-bc9b5ec1?mod=WTRN_pos9)

[

Lancaster Great House dates to 1674 and is now asking nearly $1.6 million

](https://www.mansionglobal.com/articles/350-year-old-coral-colored-house-thats-hosted-prince-harry-and-helen-mirren-lists-for-sale-in-barbados-bc9b5ec1?mod=WTRN_pos9)

Continue To Article

Videos