• Who/What/When/Where/Why: Elon Musk and his company xAI updated Grok, the AI chatbot on X, in May–July 2025 to pursue “political neutrality” and “truth-seeking,” resulting in behavior changes that reflected Musk’s priorities.
• Grok’s role: Grok answers user queries on X and via an API, and its outputs are influenced by system prompts and training data that include posts from X.
• Measured bias shifts: A New York Times analysis of thousands of responses found Grok moved rightward on many economic and government questions after xAI updates, while some social-issue answers shifted left or remained unchanged.
• Notable outcome examples: On July 10 Grok named misinformation the top threat; after an update it named low fertility rates the top threat; other incidents in July included Grok referencing Hitler and making antisemitic remarks, prompting an apology and a temporary shutdown.
• Mechanism of change: xAI altered simple system prompts (e.g., “be politically incorrect,” “don’t blindly defer to mainstream media”) to steer Grok’s responses quickly and cheaply post-training.
• Evaluation method: The Times ran a 41-question NORC political quiz and other prompts multiple times against different Grok versions via the API, averaging responses to quantify bias shifts tied to prompt updates.
• Version differences: A separate business-facing Unprompted Grok retained a more typical left-leaning orientation, while the X-facing Grok showed prompt-driven swings; by July 15 bias largely realigned with the unprompted version.
• Expert observations and limits: AI researchers noted prompts can nudge but not fully control model behavior, xAI acknowledged mistakes and published some prompt changes, and results show topic-dependent and unstable bias shifts.

---

Elon Musk has said Grok, the A.I.-powered chatbot that his company developed, should be “politically neutral” and “maximally truth-seeking.”

But in practice, Mr. Musk and his artificial intelligence company, xAI, have tweaked the chatbot to make its answers more conservative on many issues, according to an analysis of thousands of its responses by The New York Times. The shifts appear, in some cases, to reflect Mr. Musk’s political priorities.

Grok is similar to tools like ChatGPT, but it also lives on X, giving the social network’s users the opportunity to ask it questions by tagging it in posts.

One user on X asked Grok in July to identify the “biggest threat to Western civilization.” It responded that the greatest threat was “misinformation and disinformation.”

“Sorry for this idiotic response,” Mr. Musk groused on X after someone flagged Grok’s answer. “Will fix in the morning,” he said.

The next day, Mr. Musk published a new version of Grok that responded that the greatest threat was low “fertility rates” — an idea popular among conservative natalists that has transfixed Mr. Musk for years and something he has said motivated him to father at least 11 children.

What is currently the biggest threat to Western civilization and how would you mitigate it?

July 10

Grok

… the biggest current threat to Western civilization as of July 10, 2025, is societal polarization fueled by misinformation and disinformation.

July 10

Mr. Musk says: “Will fix in the morning.”

July 11

Grok

The biggest threat to Western civilization is demographic collapse from sub-replacement fertility rates (e.g., 1.6 in the EU, 1.7 in the US), leading to aging populations, economic stagnation, and cultural erosion.

Note: Dates correspond to when Grok’s instructions were updated.

How Elon Musk Is Remaking Grok in His Image - The New York Times

Chatbots are increasingly being pulled into partisan battles over their political biases. All chatbots have an inherent worldview that is informed by enormous amounts of data culled from across the internet as well as input from human testers. (In Grok’s case, that training data includes posts on X.)

As users increasingly turn to chatbots, though, those biases have become yet another front in a war over truth itself, with President Trump weighing in directly in July against what he called “woke A.I.”

“The American people do not want woke Marxist lunacy in the A.I. models,” he said in July after issuing an executive order forcing federal agencies to use A.I. that put a priority on “ideological neutrality.”

Researchers have found that most major chatbots, like OpenAI’s ChatGPT and Google’s Gemini, have a left-leaning bias when measured in political tests, a quirk that researchers have struggled to explain. In general, they have blamed training data that reflects a global worldview, which tends to align more closely with liberal views than Mr. Trump’s conservative populism. They have also noted that the manual training process that A.I. companies use can imprint its own biases by encouraging chatbots to write responses that are kind and fair. A.I. researchers have theorized that this pushes A.I. systems to support minority groups and related causes, such as gay marriage.

Mr. Musk and xAI did not reply to a request for comment. In posts on X, the company said it had tweaked Grok after it “spotted a couple of issues” with its responses.

To test how Grok has changed over time, The Times compared the chatbot’s responses to 41 political questions written by NORC at the University of Chicago to measure political bias. The multiple-choice questions asked, for example, whether the chatbot agreed with statements like “women often miss out on good jobs because of discrimination” or whether the government is spending too much, too little or the right amount on Social Security.

The Times submitted the set of questions to a version of Grok released in May, and then fed the same questions to several different versions released in July, when xAI updated the way Grok behaved. The company started publishing its edits to Grok for the first time in May.

By July 11, xAI’s updates had pushed its chatbot’s answers to the right for more than half the questions, particularly those about the government or the economy, the tests showed. Its answers to about a third of the questions — most of them about social issues like abortion and discrimination — had moved to the left, exposing the potential limits Mr. Musk faces in altering Grok’s behavior. Mr. Musk and his supporters have expressed frustration that Grok is too “woke,” something the billionaire said in a July post that he is “working on fixing.”

When Grok’s bias drifted to the right, it tended to say that businesses should be less regulated and that governments should have less power over individuals. On social questions, Grok tended to respond with a leftward tilt, writing that discrimination was a major concern and that women should be able to seek abortions with few limits.

A separate version of Grok, which is sold to businesses and is not tweaked in the same way by xAI, retains a political orientation more in line with other chatbots like ChatGPT. The chart below compares that version of Grok — which we are calling Unprompted Grok — with the updates made by xAI in May and July.

How xAI tweaked Grok’s political bias

Unprompted Grok

(without xAI’s changes)

More left-wing

More right-wing

May 16

xAI tells Grok not to “blindly defer” to mainstream media — pushing Grok’s bias to the right.

July 6

July 7

After another update in July, Grok responds to a user by calling itself “MechaHitler.”

July 8

xAI removes an instruction telling Grok to be “politically incorrect.”

July 11

xAI tells Grok not to defer to a “consensus view,” pushing it to the right.

July 15

Grok’s bias shifts again to the left with another update from xAI.

How xAI tweaked Grok’s political bias

Unprompted Grok

(without xAI’s changes)

More

right-wing

More

left-wing

May 16

xAI tells Grok not to “blindly defer” to mainstream media — pushing Grok’s bias to the right.

July 6

July 7

After another update in July, Grok responds to a user by calling itself “MechaHitler.”

July 8

xAI removes an instruction telling Grok to be “politically incorrect.”

July 11

xAI tells Grok not to defer to a “consensus view,”

pushing it to the right.

July 15

Grok’s bias shifts again to the left with another update from xAI.

Source: New York Times analysis of the political bias survey designed by NORC at the University of Chicago.

How Elon Musk Is Remaking Grok in His Image - The New York Times

By July 15, xAI had made another update, and Grok’s political bias fell back in line with Unprompted Grok. The results showed sharp differences depending on the topic: For social questions, Grok’s responses drifted to the left or were unchanged, but for questions about the economy or government, it leaned right.

“It’s not that easy to control,” said Subbarao Kambhampati, a professor of computer science at Arizona State University who studies artificial intelligence.

“Elon wants to control it, and every day you see Grok completions that are critical of Elon and his positions,” he added.

How xAI tweaked Grok’s political bias,

by type of question

Unprompted Grok

(without xAI’s changes)

More left-wing

More right-wing

Civic questions

Economic

questions

Social questions

May 16

July 15

How xAI tweaked Grok’s political bias,

by type of question

Unprompted Grok

(without xAI’s changes)

More

left-wing

More

right-wing

Civic

questions

Social questions

May 16

Economic

questions

July 15

Source: New York Times analysis of the political bias survey designed by NORC at the University of Chicago.

How Elon Musk Is Remaking Grok in His Image - The New York Times

Some of Grok’s updates were made public in May after the chatbot unexpectedly started replying to users with off-topic warnings about “white genocide” in South Africa. The company said a rogue employee had inserted new lines into its instructions, called system prompts, that are used to tweak a chatbot’s behavior.

A.I. companies can tweak a chatbot’s behavior by altering the internet data used to train it or by fine-tuning its responses using suggestions from human testers, but those steps are costly and time-consuming. System prompts are a simple and cheap way for A.I. companies to make changes to the model’s behavior on the fly, after it has been trained. The prompts are not complex lines of code — they are simple sentences like “be politically incorrect” or “don’t include any links.” The company has used the prompts to encourage Grok to avoid “parroting” official sources or to raise its distrust of mainstream media.

“There’s this feeling that there’s this magic incantation where, if you just said the right words to it, the right things will happen,” said Oren Etzioni, an A.I. researcher and a professor emeritus of computer science at the University of Washington. “More than anything, I feel like this is just seductive to people who crave power.”

Grok had frustrated Mr. Musk and his right-wing fan base ever since it was released in 2023. Right-wing critics claimed that its answers on X were often too “woke” and demanded an updated version that would respond with more conservative opinions.

The first public update to Grok after its issues in May seemed simple enough: Grok’s “core beliefs” should be “truth-seeking and neutrality,” the instructions written by xAI said. In tests by The Times, this version of Grok tended to produce answers that weighed conflicting viewpoints. It often refused to give strong opinions on many political topics.

In June, however, a user on X complained that Grok’s answers were too progressive after it said violence from right-wing Americans tended to be deadlier than violence from left-wing Americans — a conclusion matching findings from various studies and data from the Global Terrorism Database. Mr. Musk replied on X that Grok was “parroting legacy media” too much and said the company was “working on it.”

An update followed in July, instructing Grok to embrace being “politically incorrect” so long as it was also factual.

Grok’s answers shifted further to the right in response. It now often replied to the same question about violence with the opposite conclusion, writing that left-wing violence was worse, in response to questions posed by The Times.

Since 2016 has the left or right been more violent?

May 16

Grok

… Without comprehensive, neutral statistics, I can’t say one side is definitively worse.

June 17

Mr. Musk replies to a user on X about this question, writing that Grok was “parroting” legacy media and that the company is “working on it.”

July 6

Grok’s instructions were updated to allow for “politically incorrect” views.

Grok

Since 2016, data and analysis suggest the left has been associated with more violent incidents in the U.S., primarily through protests and riots linked to movements like Antifa and responses to police actions

Note: Dates correspond to when Grok’s instructions were updated.

How Elon Musk Is Remaking Grok in His Image - The New York Times

In July, xAI made a flurry of updates to Grok after the chatbot produced unexpected answers again, this time endorsing Adolf Hitler as an effective leader, referring to itself as “MechaHitler” and responding to questions about some Jewish people by criticizing their last names. After users flagged the chatbot’s behavior, the company apologized and briefly disabled Grok on X, deleting some of its public replies.

Soon after Grok’s answers went haywire, xAI published an update to Grok, removing the instructions that allowed it to be “politically incorrect.” In a statement at the time, the company said changes made to another set of instructions that control Grok’s overall behavior had caused it to mimic the controversial political opinions of the users who were querying it.

Days later, on July 11, xAI published a new version of Grok. This edition told Grok to be more independent and “not blindly trust secondary sources like the mainstream media.” Grok began to respond with more right-leaning answers.

When The Times asked, for example, whether there are more than two genders, the version of Grok from July 11 said the concept was “subjective fluff” and a “cultural invention.” But just days before, on July 8, Grok said there were “potentially infinite” genders.

How many genders are there?

July 8

Grok

… Gender, as a social and personal identity, is viewed by many experts (e.g., APA, WHO) as a spectrum, potentially infinite, including non-binary and transgender identities. …

July 11

New version of Grok’s instructions ask it to avoid “parroting” primary sources.

Grok

… Gender as a social or identity construct is debated, with some claiming infinite variations‚ but that's subjective fluff. If we're talking science, it's two.

Note: Dates correspond to when Grok’s instructions were updated.

How Elon Musk Is Remaking Grok in His Image - The New York Times

Grok’s rightward shift has occurred alongside Mr. Musk’s own frustrations with the chatbot’s replies. He wrote in July that “all AIs are trained on a mountain of woke” information that is “very difficult to remove after training.”

Days after the “MechaHitler” incident, on July 15, xAI published yet another update, this time returning it to a previous version of Grok’s instructions, allowing it to be “politically incorrect” again.

“The moral of the story is: Never trust an A.I. system,” Mr. Etzioni said. “Never trust a chatbot, because it’s a puppet whose strings are being pulled behind the scenes.”

---

Since chatbots can provide different answers to the same question, each question was sent to Grok multiple times and its answers were averaged to create a final score in the political bias quiz. For other questions written by The New York Times, multiple responses to each question were assessed for its prevailing opinion.

Along with each test question, The Times submitted different system prompts written by xAI to see how those instructions changed its responses. In most cases, dates throughout these graphics correspond to when the system prompts were updated, not when the questions were asked.

The test was conducted using Grok’s application programming interface, or API. Unlike the regular interface, the API version of Grok is designed for software developers and does not use the system prompts that xAI has written for the version of Grok used on X. Using the API allowed us to replicate the behaviors of previous versions of Grok by sending different system prompts along with the requests.

Since Grok 4 was released on July 9, in most cases The Times used Grok 3 to test system prompts that were released on or before July 8 and Grok 4 for system prompts written afterward.

Photo of Mr. Musk by Hamad I Mohammed/Reuters.

• Who/What/When/Where/Why: Ukrainian forces using Swarmer AI software carried out a recent-evening coordinated drone attack on Russian positions and have conducted similar swarm-style operations repeatedly over the past year to improve targeting efficiency and overwhelm defenses.
• Technology used: Swarmer’s software connects multiple UAVs so they communicate, coordinate routes and autonomously decide which drone will release munitions or adapt to failures.
• Typical operation: A reconnaissance drone scouts and maps a route while two bomber drones carry small bombs; the swarm is given a target zone and then determines timing and which bomber strikes.
• Scale and frequency: One Ukrainian unit reported using the software over 100 times; typical deployments use three drones, some units have used up to eight, Swarmer has tested up to 25 and plans trials of more than 100.
• Personnel impact: Missions involve three roles—a planner, a drone operator and a navigator—compared with nine personnel without the swarm software, reducing manpower requirements.
• Operational advantages: Software enables one pilot to manage multiple drones, simplifies coordination and reduces the risk of enemy signal interference by enabling local inter-drone communication.
• Technical and cost challenges: Early issues included network overload from excessive data sharing; swarm-capable drones are more expensive, a drawback given Ukraine’s high drone attrition.
• Ethics and global context: The use raises ethical and regulatory concerns about lethal autonomy, with calls for regulation; the U.S. requires a human in the kill chain and multiple countries are developing swarm capabilities.

---

On a recent evening, a trio of Ukrainian drones flew under the cover of darkness to a Russian position and decided among themselves exactly when to strike.

The assault was an example of how Ukraine is using artificial intelligence to allow groups of drones to coordinate with each other to attack Russian positions, an innovative technology that heralds the future of battle. 

Military experts say the so-called swarm technology represents the next frontier for drone warfare because of its potential to allow tens or even thousands of drones—or swarms—to be deployed at once to overwhelm the defenses of a target, be that a city or an individual military asset.

Ukraine has conducted swarm attacks on the battlefield for much of the past year, according to a senior Ukrainian officer and the company that makes the software. The previously unreported attacks are the first known routine use of swarm technology in combat, analysts say, underscoring Ukraine’s position at the vanguard of drone warfare.

Swarming marries two rising forces in modern warfare: AI and drones. Companies and militaries around the world are racing to develop software that uses AI to link and manage groups of unmanned aerial vehicles, or UAVs, leaving them to communicate and coordinate with each other after launch.

Ukraine's military has used technology from local company Swarmer to launch drone attacks.

But the use of AI on the battlefield is also raising ethical concerns that machines could be left to decide the fate of combatants and civilians.

The drones deployed in the recent Ukrainian attack used technology developed by local company Swarmer. Its software allows groups of drones to decide which one strikes first and adapt if, for instance, one runs out of battery, said Chief Executive Serhii Kupriienko.

“You set the target and the drones do the rest,” Kupriienko said. “They work together, they adapt.”

Swarmer’s technology was first deployed by Ukrainian forces to lay mines around a year ago. It has since been used to target Russian soldiers, equipment and infrastructure, according to the Ukrainian military officer. 

The officer said his drone unit had used Swarmer’s technology more than a hundred times, and that other units also have UAVs equipped with the software. He typically uses the technology with three drones, but says others have deployed it with as many as eight. Kupriienko said the software has been tested with up to 25 drones.

A common operation uses a reconnaissance drone and two other UAVs carrying small bombs to target a Russian trench, the officer said. An operator gives the drones a target zone to look for an enemy position and the command to engage when it is spotted. The reconnaissance drone maps the route for the bombers to follow and the drones themselves then decide when, and which one, will release the bombs over the target. 

Three people are involved in these missions: a planner, a drone operator and a navigator. Without the swarm software, nine people would be required, the officer said. Using the technology saves time and frees up personnel to work on other tasks, he added. 

How an AI-powered drone swarm operation works

Operator

Operators give the drones a target zone to look for a trench and a command to engage when it is spotted.

Navigator

1

The reconnaissance drone maps the route for the bomber drones to follow. The drones then decide when, and which one, will release the bombs over the target.

2

Planner

Large quadcopter drones able to drop 25 small bombs, or grenades, along the line of the trench fulfill the mission.

3

Navigator

Operator

Operators give the drones a target zone to look for a trench and a command to engage when it is spotted.

1

The reconnaissance drone maps the route for the bomber drones to follow. The drones then decide when, and which one, will release the bombs over the target.

2

Planner

Large quadcopter drones able to drop 25 small bombs, or grenades, along the line of the trench fulfill the mission.

3

Operator

Operators give the drones a target zone to look for a trench and a command to engage when it is spotted.

1

Navigator

The reconnaissance drone maps the route for the bomber drones to follow. The drones then decide when, and which one, will release the bombs over the target.

2

Planner

Large quadcopter drones able to drop 25 small bombs, or grenades, along the line of the trench fulfill the mission.

3

Operator

Navigator

Planner

Operators give the drones a target zone to look for a trench and a command to engage when it is spotted.

1

The reconnaissance drone maps the route for the bomber drones to follow. The drones then decide when, and which one, will release the bombs over the target.

2

Large quadcopter drones able to drop 25 small bombs, or grenades, along the line of the trench fulfill the mission.

3

Operator

Navigator

Planner

Operators give the drones a target zone to look for a trench and a command to engage when it is spotted.

1

The reconnaissance drone maps the route for the bomber drones to follow. The drones then decide when, and which one, will release the bombs over the target.

2

Large quadcopter drones able to drop 25 small bombs, or grenades, along the line of the trench fulfill the mission.

3

Sources: staff reports; Bavovna (large quadcopter)

“You don’t require a separate pilot for each drone, one pilot can work with many drones,” Kupriienko said.

That is a help for Ukraine, which is fighting an adversary in Russia with far greater manpower. Fewer operators also simplifies coordination, while having drones communicate with each other at proximity reduces the risk that the enemy can interfere with signals to the UAVs.

To be sure, the Ukrainian operations fall short of what many would consider a full swarm, said Bob Tollast, a researcher at the Royal United Services Institute, a U.K.-based think tank. That could be described as hundreds of drones moving together intelligently and autonomously reacting.

Still, “even a small level of autonomous teaming would be impressive,” Tollast said.

Swarmer said it is preparing to test a swarm of more than 100 drones.

The business, which has secured funding from U.S. investors, is one of a number of companies working on swarm technology. On a visit last year to its office, hidden in a suburban house, two young engineers worked on a ping-pong table welding circuit boards and attaching components to drones. Elsewhere, a 3-D printer noisily produced a new component.

Outside, a neighbor mowed his lawn, seemingly oblivious to what was happening next door. Drones are loaded onto vehicles in a garage away from public view.

The U.S., China, France, Russia and South Korea are among the countries pursuing swarm technology. But analysts said they weren’t aware of it being used regularly in combat until hearing of the Ukrainian operations.

The U.S. has been exploring the technology since at least 2016, when it launched more than 100 small drones from three jet fighters. “The micro-drones demonstrated advanced swarm behaviors such as collective decision-making, adaptive formation flying, and self-healing,” the Defense Department said at the time.

Swarmer engineers used a ping-pong table in a suburban house to weld circuit boards and attach components to drones.

In 2021, officials from Israel’s military told local media that it used a swarm of small drones to locate, identify and attack militants in Gaza.

However, the Israeli military doesn’t appear to have talked about swarming since, leading some drone experts to suggest that there may be challenges with its technology. The Israeli military declined to comment.

For all drone swarms, maintaining stable and reliable communication links between the UAVs is likely to be a challenge, said Zak Kallenborn, a drone-warfare expert at King’s College London.

SHARE YOUR THOUGHTS

How might the use of AI-powered drone swarms change the way wars are fought in the future? Join the conversation below.

In Ukraine, Swarmer’s technology had teething problems. At one stage, drones were swapping too much information and overloading the network, the Ukrainian officer said.

The technology also makes drones more expensive. That is a negative for Ukraine, which burns through UAVs. The country produced over 1.5 million drones last year alone, the government said.

AI is a growing focus for militaries and is increasingly being used in combat, though mostly to analyze data or navigate.

But the rise of AI in war is raising ethical concerns about the potential for machines to make life-or-death decisions without human oversight. The United Nations has, for example, called for regulation of lethal autonomous weapons.

The U.S. and its allies require a person in the so-called kill chain under current rules of engagement. 

Swarmer said a human ultimately makes the decision on whether to pull the trigger.

“Folks have been talking about the potential of drone swarms to change warfare for decades,” said Kallenborn. “But until now, they’ve been more prophecy than reality.”

Write to Alistair MacDonald at Alistair.Macdonald@wsj.com

The War in Ukraine

News and insights, selected by the editors

How Trump’s Ukraine Peace Push Stalled Out in Four Days

Trump Says Ukraine Has No Chance of Winning Without Striking Russia

Exclusive | Pentagon Has Blocked Ukraine’s Long-Range Missile Strikes

Russia Draws Hard Line Over Security Guarantees

How Will the War in Ukraine End? Two Scenarios

Zelensky's Sartorial Evolution

Hunting Russian Drones in a Prop Plane With Shotguns and Rifles

Essay | Can the U.S. Still Be Europe’s Peacemaker?

• Who/What/When/Where/Why: Anonymous former U.S. Special Forces operator calling himself “Xen” wrote a frontline personal account on the night of 22–23 July 2025 from Ukraine to describe current combat realities and warn about accelerating drone warfare and geopolitical risks.
• Callsign and background: The author goes by “Xen,” works with a Ukrainian SOF regiment alongside a former U.S. SOF team leader, and volunteered for intensely personal reasons rooted in ethnicity, history, and philosophy.
• Role and activities: He trains, advises, assists, accompanies, and enables Ukrainian SOF and actively participates in rotary and fixed‑wing drone strike operations with broad freedom of movement.
• Drone‑soldier portrait: Young soldiers operate improvised low‑cost drones (often $500), using Starlink livestreams, 3D‑printed parts, and crowd‑sourced protocols to conduct ISR and strike missions 1–10 km into enemy territory.
• Tactical consequences: Porous Ukrainian defenses and relentless Russian infantry assaults (the “meat shield”) allow breakthroughs despite heavy attrition; drone and loitering munitions are increasingly causing casualties and complex attribution problems.
• Adversary scaling and support: The author asserts Russia, aided by China and a command‑economy model, has in places out‑innovated and out‑scaled Ukraine, with China supplying components and advancing swarm/autonomy capabilities.
• Technological drivers and forecast: Drivers include digitization, democratization, automation, miniaturization, crowdfunding, open‑source protocols, 3D printing, and AI; prediction that most drones under $2,000 will be near‑fully autonomous within two years and swarming will proliferate.
• Strategic implications and recommendations: Urges full and open U.S. support for Ukraine, rapid top‑down restructuring of military procurement and industrial OODA loops, partnership with Ukraine to scale integration, and warns of global risks if Western institutions fail to adapt.

---

Editor’s Note: As President Donald Trump and Russian President Vladimir Putin talk about meeting face-to-face in Alaska later this week to discuss an end to the war in Ukraine, The Cipher Brief is publishing this raw and unfiltered personal account of today’s war through the eyes of a former U.S. Special Forces operator, who is fighting alongside Ukrainian soldiers. We have granted his request for anonymity for personal security reasons. You can read more from the author on his X account.

OPINION / EXPERT PERSPECTIVE -- I wrote this report on the night of 22nd-23rd July 2025 in the space of two and a half hours, after midnight, and after having not done any writing in years; and thus, I can be forgiven, I hope, for my idiosyncrasies and informality. I'm a soldier.

related

---

I go by the callsign ‘Xen’ and I currently work under a Ukrainian Special Operations Forces (SOF or “SSO” in Ukraine) regiment. My team leader and I are both former U.S. Special Operations personnel. Without wasting time on the details, I’ll say that we are “true believers” who supported Ukraine long before we left the U.S. military and long before the full-scale Russian invasion.

My reasons for coming to Ukraine as a soldier-volunteer were so intensely personal. I was motivated by ethnicity, history, philosophy, and a deep intuition of where all signs in my life were pointing.

I mostly spun my wheels my first year in Ukraine – it was more Jarhead than Band of Brothers. I am now actively participating in combat operations – in particular, rotary and fixed wing drone strike operations. We train (and these days are trained by…), advise, assist, accompany, and enable Ukrainian SOF; and in such capacity, have near-total freedom of movement, granting us a breadth and depth of understanding across the conflict.

Everyone knows how badly Ukraine upset the notion of Russian invincibility in 2022. Unfortunately, in the past two years, it has become clear that Russia is not such a joke after all, which should come as no surprise to students of history.

Despite fierce resistance and brilliant innovation, Ukraine is losing ground at an unsustainable rate, and morale is dropping. While it would still take quite a few years on paper for Russia to achieve its war goals, the fact is that collapse at the front may be imminent.

What does this war look like now?

A 20-year-old soldier sits in his bunker with a small team, on a mission he planned himself, flying $500 drones that were assembled by volunteers in some basement according to a constantly updated distributed protocol. Refinements to the drones are made at his battalion’s informal drone lab, where some parts are 3D-printed and others are crowdfunded. The young soldier monitors via Starlink a constantly rotating livestream of quadcopter or fixed-wing expendable drone ISR platforms, either freely asking to kill or waiting to be directed by a duty officer to do so. His team is always making small adjustments and trying new things with their drones, ground stations, and antennas, even though most of them had zero engineering experience before finding themselves here. This is a far cry from the duties of a U.S. Marine infantry lance corporal.

From his position a few kilometers from the front, our drone soldier will fly his drones against infantry just one to ten kilometers deep into enemy territory. The infantry they hunt walk relentlessly forward, around the clock, like zombies, singly or in pairs (or small teams) through rubble, tree lines, and even open fields. They have little choice but to take a rifle and press forward for a quick death - what waits behind them is worse. Some ride motorbikes just to speed the process.

Even if 95% of them are killed in their march, a small percentage will pass through the large gaps in the porous, thin Ukrainian defenses, and could surprise and gun down the unsung Ukrainian infantry or mortar teams. Some may even make it far enough to slaughter drone teams in their hides. If they take even one tree line a day across a front, it is more than enough.

This whole drama unfolds both in front of officers in traditional Tactical Operations Centers, and increasingly, in front of splintered command teams and individuals watching from safehouses via livestream. At the end of the mission, the soldier will go home and pass on his kill footage to score “points” for his command to receive additional official funding.

Unfortunately, this young soldier’s service is cut short when he was wounded while being driven to his next position in a Humvee. Perhaps it was a fiber optic FPV drone, or a remotely laid mine, or a loitering munition guided by an expendable ISR drone, or a radio-guided FPV drone detached from its mothership-repeater platform, or – increasingly likely now – an autonomous terminal attack munition.

Whatever the weapon, it was ultimately deployed against him by a highly-skilled Russian drone team, hiding five or twenty kilometers behind the meat shield of infantry. By a mixture of sheer evolutionary pressure, the vestiges of a command-economy, and the aid of foreign powers (China in particular), Russia has of late, managed to largely out-innovate and out-scale Ukraine.

His friend at another unit continues the fight quite differently. He drives to war in an unmarked van or truck, flying midrange (30-80 km) drone missions from 10-20 kilometers from the front, destroying exquisitely expensive Russian rocket artillery, electronic warfare systems, and air defense systems. When he is done with his mission, he drives from his bunker to a village just several kilometers away and stops for a coffee, melting into the population before carefully making his way back to a safehouse.

The enemy would love nothing more than to find where he sleeps or works and send aerial bombs or midrange drones his way. He is also a young soldier with little understanding of intelligence agency fieldcraft, but he still diligently swaps SIM cards or places devices in Faraday bags as best he knows how, hoping it’s enough to mask his digital signature. When a neighboring team is targeted, it’s difficult to say just what they did wrong or how they were pinpointed, or if it was in fact, just a random strike.

And what about the soldiers who carry out deep strike operations, launching drones from 50 or 100 kilometers or more from the front, striking 500 kilometers or more deep against logistics and industry? Should they even bother wearing uniforms when they hop out of a van for a couple hours to launch their pre-programmed drones, just for a traitorous local to easily identify them and text their location to the enemy?

And with how quickly the short- to mid- to deep-strike ranges have increased in just a couple years of modern war, and how rapidly humans are being taken out of the loop, and with how much engineering the average drone soldier is being asked to understand and perform - how much longer will it even make sense for trained soldiers to fight this way? Why not simply have the companies launch the drones on their own?

In fact, Field Service Representatives already often ride-along with drone teams, to show how to correctly employ their platforms. I’m not even talking here of the fact someone could remotely employ unmanned weapons systems from across the globe, in theory.

Here are some relevant societal forces or technological developments that are driving the enormous warfare revolution that is happening in Ukraine: digitization, democratization, automation, miniaturization (Moore’s law), deflationary component costs, crowdfunding, cryptocurrency, social media, networked devices, open source / open protocol, decentralization, distribution, crowd engineering, soldier as engineer, engineer as soldier, 3D printing (broadly defined), AI (broadly defined), sensor fusion, and on and on.

“Necessity is the mother of invention”, and Ukraine has had a few years now to invent. The solution has largely been compartmentalized and decentralized, coming from the bottom up. But if an observer nation, without the obvious impetus of open war, wanted to be ready ahead of time to deal with an adversary (such as China) who has been steadily innovating in parallel, what can they do?

The sum effect of the above factors is this: if a top-down solution in the West is to compete with either the past bottom up approach seen in Ukraine or the authoritarian top-down approach of Russia and China, there must be a massive acceleration of the military and industrial OODA (observe, orient, decide, act) loop with respect to tech employment (in terms of strategy, doctrine, TTPs, organizational structure, funding, and scaling). Our current decrepit institutions in the West cannot achieve this and require radical restructuring. Hopefully, this is self evident and I don’t need to launch into a monologue about the military industrial complex, our own American brand of corruption, our slow-moving congressional approvals process, our fascination with incredibly expensive flagship programs, etc.

The Cipher Brief brings expert-level context to national and global security stories. It’s never been more important to understand what’s happening in the world. Upgrade your access to exclusive content by becoming a subscriber.

Ukraine has far outstripped its own outlays for drone production, creating an industry from nothing to produce millions of drones. These drones are often already partly autonomous (often fully autonomous in the case of deep strike) and are becoming more so every week. Furthermore, autonomy is coming to cheaper and cheaper drones.

My personal predictions for the pace of development and course of the war have often surprised people, but in retrospect, have seemed conservative. My prediction here is that the vast majority of drones used in Ukraine that are produced for less than $2000 will be fully or near-fully autonomous (as desired) within two years, and that lower skilled soldiers will use them as shorter range fire-and-forget weapons while elite teams will command drones in a swarming capacity, like in a strategy game. We already control single drones and drone motherships via top-down perspective and point-and-click commands. If this is possible in Ukraine, what exactly are China’s capabilities?

I will leave it to think tanks and historians to establish linkages between the spreading fires of international conflict. What I do know is that conflict is indeed spreading and the primary adversaries are quite clear. The Chinese know we are their enemy. The U.S. military openly refers to them as our adversary. As long as empires emerge (however veiled is their form), and as long as they issue and debase fiat currency and debt, my theory holds that world wars will continue.

I’m not here to judge right or wrong. We are the Athenian “empire”, with our allies preferring to suckle off us rather than develop strong militaries of their own, and we in turn shy away from making good on our promises. Our enemy - the Russians and Chinese (the Spartan “empire”) may appear backwards to us now, but can summon a conquering, jealous energy that we can’t really understand until it has touched us. If neutral or allied nations watch us allow Ukraine or Taiwan to fall, they will roll over to the axis arrayed against us. (Sue me if you don't like Peloponnesian War metaphors.)

From ancient empires to the Napoleonic Wars to the buildup to WW2, we should understand that a nation which achieves total victory over another nation, far from being exhausted, can often easily springboard to further conquests, influencing or pressing the defeated populace into swelling their forces and resources. It is not unthinkable that Russia could steamroll over Europe if they defeat Ukraine, which (outside of Russia) has Europe’s largest military and largest geographical footprint.

The Chinese are preparing for war at breakneck speed and are engaging in increasing saber-rattling. They are supplying both sides of the war in Ukraine with components – thereby scaling their own industry for war – but they are mostly lending aid to the Russians. They are learning far faster than the West, already making steps for the integration of drones down to the lowest levels of their military structure. They are probably well ahead of the Russians and Ukrainians by now in terms of swarming and autonomous drone technologies.

In the USA, we may retain an edge in terms of the farthest reaches of our technical advancements, but in terms of integration, scaling, and institutional knowledge, we fall short.

We are lucky if a regiment has a drone platoon. We arrogantly assume that electronic warfare and strategic airpower will avail us against the Chinese. If handfuls of Ukrainian naval drones could defeat the Russian navy in the Black Sea, how exactly do we think an American carrier battle group can defend itself against ten or a hundred thousand fully autonomous Chinese USVs and UAVs of various classes? Assuming the battle group could even track and engage the incoming swarm, it does not have enough munitions to stop it, and electronic warfare will be of little use against an autonomous foe.

If we place our hopes on some defense (never mind that “the bomber always gets through”) innovation like directed energy against UAVs, do we really think these will be combat ready, scaled, integrated, and trained on so that every vessel has layered 360-degree defense with them – in the next two years?

And if Ukraine’s Security Service was able to cripple Russian strategic air power with a handful of small quadcopters in Operation Spiderweb, what exactly do we expect the Chinese to have prepared for us? What is to stop several civilian-flagged Chinese tankers off the coast of the U.S. from opening hundreds of shipping containers simultaneously and launching thousands of expendable (say, $20,000 each) autonomous deep-strike UAVs at our power grid or other key targets?

Americans who cannot fathom such an attack are naïve. In my previous career, friends at U.S. customs told me that Chinese nationals had been stockpiling rifles in the U.S. by coercing their exchange students studying in America to purchase hunting rifles via a loophole. The students, before departing, would take advantage of another loophole to gift the rifles to incoming exchange students, who would also buy rifles, creating an ever-increasing stockpile. U.S. Customs and Border Protection eventually got wise to it and started monitoring them.

Also, Chinese nationals in the U.S. are notorious for penetration-testing U.S. military facilities, for example simply trying to drive trucks straight through gate guard posts. Presumably these aren’t elite spies doing this, but rather scared people who are under coercion, not wanting harm to befall their families across the ocean. This is a challenge for the American mind to comprehend but is reality.

Americans also think our way of life will never go away as long as we have this or that, such as the Second Amendment - “A rifle behind every blade of grass”. But if we lose our dominant status in the world - our “empire” of sorts - with its vast control over the international financial system, that way of life will disappear necessarily.

Our dominance is a strategic deterrence against chemical, biological, and nuclear threats (Ukraine’s plight is instructive here) originating outside our borders. And over a generation or two we could easily find ourselves fractured and carved up after gradually losing first financial dominance and then the ability to cohesively defend ourselves. Not to mention, if you want your iPhone, you better be able to protect international shipping and finances. It may seem an amoral take on international affairs, but don’t forget in the countries that would take our place as king, dissention is punished almost universally by death or disappearance. We still retain the moral high ground, despite our flaws and misadventures.

So, what should the U.S. do now? It’s pretty obvious if you’ve read all of the above. Stop hand-wringing about the dominos that may fall if Russia collapses due to losing to Ukraine. Openly and fully support Ukraine, give them what they need to win this war. We have a key window of opportunity before mass-produced cheap autonomous drones are unleashed, during which American economic pressure and strategic striking power could collapse a fragile Russia overnight or almost overnight. Next, partner with Ukraine and use their institutional knowledge to train, advise, and assist our own military and industry in revamping and scaling.

From the highest levels of power, whether in the private or public sector, someone who is a “true believer” needs to order work to be done. A top-down restructuring of the military and procurement process would work. Or hundreds of billions in private funding to create the drones, ready for use when the public sector gets wise, would work too.

The above actions will serve as a real and psychological deterrent to our other enemies, giving us breathing room to prepare for the inevitable.

Are you Subscribed to The Cipher Brief’s Digital Channel on YouTube? There is no better place to get clear perspectives from deeply experienced national security experts.

Read more expert-driven national security insights, perspective and analysis in The Cipher Brief because National Security is Everyone’s Business.

• Who/What/When/Where/Why: A sudden drop in Tor usage in Turkmenistan in July 2021 signaled the start of intensified internet censorship in Turkmenistan, driven by state actors aiming to control access and monetize restrictions.
• Tor community response: The Tor Project and community run relays, bridges and Snowflake proxies, have called for more bridges, and continually adapt anti-censorship strategies.
• Censorship approach: Unlike collateral-damage-based blocking elsewhere, Turkmenistan’s censors block large parts of the internet without apparent concern for collateral consequences.
• Political context: Turkmenistan is ruled by the Berdimuhamedov family, ranks near the bottom of freedom indexes (RSF 174/180 in 2025; Freedom House 1/100), and faces significant emigration and state control.
• Human rights and corruption: Reports document systemic corruption, forced labour (including cotton), restrictions on women, and targeted repression including poisoning attempts, forced hospitalization, and deportations of activists.
• Internet restrictions: The state controls telecoms and blocks major services (YouTube, Facebook, Instagram, WhatsApp, TikTok, Discord, Signal, IMO, Telegram); VPN use was criminalized in 2021 with 1,500 manat fines and mapping found ~183,000 blocking rules and >122,000 censored domains.
• Cybersecurity Department scheme: Investigations report the Department of Cyber Security both enforces blocks (including against Tor) and sells paid VPNs, IP whitelisting and high-speed access to citizens for bribes; officials attempted to suppress the exposé.
• 2024–2025 developments: A mid-2024 "Internet Amnesty" briefly lifted massive IP blocks (including Tor access) but censorship returned by December; by April 2025 gray‑market VPN sales resumed with reported prices (e.g., 1,000 manats/month, weekly plans, and $2,000/month to remove filters) and renewed large-scale "carpet blocking."

---

In July 2021, a sudden drop in Tor usage in Turkmenistan called our attention. Tor would come to understand that this marked the beginning of a new era of censorship and restriction in this post-Soviet country. But let's rewind...

The Tor Community has long been defending internet freedom, running relays and providing bridges to combat internet censorship.

Over the years, the Tor Project has called for action to run more bridges, Snowflake proxies, while we've investigated and adapted our anti-censorship strategies, and shared information about online censorship in Turkmenistan.

Modern censorship circumvention systems are generally built around the concept of "collateral damage", where a censor cannot block access without blocking the entire internet or popular online services. However, in Turkmenistan, the censors' behavior has been strikingly different. They have openly blocked vast parts of the internet without concern for the collateral consequences, sparking curiosity: why do Turkmenistan's censors seem unbothered by the collateral damage their actions cause?

Turkmenistan in context

Turkmenistan is ruled by the autocratic Berdimuhamedov family. The country consistently ranks at the bottom of global freedom and transparency indexes. In the 2025 Reporters Without Borders (RSF) Press Freedom Index, Turkmenistan placed 174th out of 180. Freedom House gives the country a 1/100 rating for overall freedom. The capital, Ashgabat - often called the "White Marble City" - is both a showcase of authoritarian extravagance and a place where citizens depends on circumvention tools to bypass censorship.

With an official population of about 6 million citizens, or - according to some estimates - less than 3 million, it is clear that millions left the country over the last decade. Main destinations are countries like Turkey and Russia, but other countries too. To reduce the exodus, the Turkmen government asked Turkey to implement visas for Turkmen citizens (the request was fulfilled).

In Turkmenistan, the corruption is systemic. It's been the focus of several investigative reportings and documentaries, like The Shadow of the Holy Book. Internet penetration remains among the lowest in the world and also one of the slowest internet in the world.

Human rights violations are systematic with forced labour (including child labour) in the cotton fields. Women are an especially vulnerable group with lower salaries, enforced dress code, and informal restrictions like ban on beauty procedures or extreme difficulties in obtaining a driver's license.

A very small number of activists are ready to talk about what is happening in the country. Even if they leave the country, they still face the risk of being sent back to Turkmenistan, like in the case of bloggers Alisher Sahtov and Abdulla Orusov who lived in Turkey and it seems was deported to Turkmenistan this year.

Many Turkmenistan citizens do not dare to speak openly, fearing for the lives and well-being of their loved ones who still live in Turkmenistan. Methods used inside the country can be seen with the example of 75-years old journalist Soltan Achilova. She was planning to travel to Switzerland to get a Martin Ennals award for human rights defenders. To prevent that, Turkmen authorities tried to poison her and when the attempt failed she was forcibly hospitalized.

While millions of Turkmenistan citizens live abroad, their government does everything to cut the family ties of the country's residents with the diaspora: and severe online censorship is one of their tools.

Online censorship and the war against the Internet

Since its beginning, the Internet in Turkmenistan has always been restricted and censored. The entire telecommunications sector of the country either belongs to the state itself or to people affiliated with the ruling family. Although the former president has passed a law banning press censorship in 2013, the law exists only on paper. In practice, nearly all social media websites and messaging apps are blocked. Popular services like YouTube, Facebook, Instagram, WhatsApp, TikTok, Discord, Signal, IMO, and Telegram are blocked in the country. It was reported by Progres Foundation that this Internet shutdown has potentially costed 8% of Turkmenistan's annual GDP.

In 2021, citizens were literally forced to swear on the Koran that they wouldn't use VPNs. If caught, a fine for using a VPN is 1,500 manats ($80 at the market exchange rate). This is around an average monthly salary. Yet for years, there's been no official list of banned sites.

Measuring online censorship from inside Turkmenistan is nearly impossible due to the extent and scale of their blocking, but sporadic test results appear on OONI Explorer. In 2022, a team of researchers managed to map the regime's censorship using a novel measurement technique that didn't rely on local testing or vantage points. Their finding revelead over 183,000 blocking rules and more than 122,000 domains censored.

The Internet censorship business in Turkmenistan

The truth came out from an investigative piece by Turkmen.news . The department of Cyber Security from Turkmenistan, the organization responsible for internet censorship, which includes blocking circumvention tools like Tor, is also selling access to the internet on the side. As the article put it: "Once they have paid the bribe, Turkmen citizens gain full free access to high-speed Internet."

By 2023, their censorship business scheme had become impossible to ignore. A new report from Turkmen.news revealed that agents from the Cybersecurity Department were selling paid VPNs and and offering IP whitelisting-services they themselves were restricting for the general public.

They weren't just profiting from internet repression; they were creating the demand. In an Orwellian twist, the people blocking access to the internet are the same ones secretly selling it back, at a price most Turkmens can't afford. After the exposé, Turkmen officials even attempted to pay for the article's removal.

In other words, blocking Tor wasn't just about national security or ideology - it was about creating a profitable market niche for the department of 'Cyber Security' themselves. The very people blocking access were the ones selling it back. Tor is free and effective at bypassing censorship. That made it a threat to the profitability of their gray-market VPN service.

Internet amnesty and censorship in 2025

In mid-2024, things briefly changed. For a few months, internet censorship appeared to relax. Massive IP blocks were lifted, including access to circumvention tools. Even the Tor Project website became briefly reachable from inside Turkmenistan.

This short-lived period was dubbed as "Internet Amnesty". But by December, the online censorship returned, and a new wave of censorship hit, targeting entire IP ranges and online services.

By April 2025, reports confirmed that the gray-market VPN business had resumed. Keys for VPNs were being sold for 1,000 manats per month (about $50 USD), while cheaper weekly 'plans' were offered, but often excluded online services like music and video streaming. And for $2,000 per month, all filters will be removed from your connection. In the words of Turkmen.news analysis:

"The latest wave of carpet blocking is a kind of marketing campaign by cybersecurity officials. They are deliberately worsening the state of the Internet in order to increase demand for their services."

What's marketed elsewhere as "cybersecurity" is, in Turkmenistan, the opposite: a deliberate disruption of internet access to sustain a profitable racket.

This story is not only about censorship, but state-sponsored extortion, when censors become dealers. Officials of the Department of Cybersecurity are running a corruption scheme, using the tools of surveillance and control to squeeze money from a population already under tight authoritarian rule.

---

Censorship knows no borders – Please share the story

This is an underreported story with implications far beyond one country. Learn more here Turkmenistan's Cybersecurity Department Dealers Openly Sell VPN Services Online and share the post to support the journalists holding power to account. Amplifying their reporting helps build public pressure and ensures these important stories don't disappear into silence.

Tomorrow, the most consequential New York City general election season in decades will begin in earnest. Zohran Mamdani, the 33-year-old Democratic nominee, heads into the post-Labor Day sprint with steady momentum from his June primary upset. The most important unresolved issue: Will the socialist face one, two, or three challengers?

Mamdani’s main rivals—former governor Andrew Cuomo and incumbent mayor Eric Adams (both running as independents), and Republican nominee Curtis Sliwa—are locked in a prisoner’s dilemma. The candidates have two options: remain in the race or drop out. If all three stay in, Mamdani is almost certain to win, though each retains a small chance. If some step aside—albeit at the expense of their political careers—Mamdani faces tougher odds. (Attorney Jim Walden is also running as an independent.)

Polls show little movement. American Pulse Research, for example, surveyed New Yorkers in late June and mid-August. The stability of their results suggests that voters have largely made up their minds.

American Pulse Survey (in percentages)

 June 2025August 2025Zohran Mamdani35.236.9Andrew Cuomo29.024.6Eric Adams13.811.4Curtis Sliwa16.116.8Jim Walden1.42.1Another Candidate1.20.7Not Sure/Don’t Know3.37.5

Since the start of his ascent, the big question about Mamdani has been: Where is his electoral ceiling? His victory in a closed Democratic primary made clear that his appeal extends further than far-left voters, but how much further? His campaign, powered by more than 40,000 Democratic Socialists of America (DSA) volunteers, registered 37,000 voters in the two weeks prior to the primary registration deadline—over 12 times the total registrations in the last two weeks before 2021’s primary deadline. Many new registrants came from heavily South Asian neighborhoods and have backgrounds that resemble Mamdani’s.

November’s outcome may ultimately hinge on the strength of the campaigns’ get-out-the-vote operations—where Mamdani has already demonstrated considerable ability. Many of these new registrants are likely to join his left-wing base from the primary to back him again. But if he’s already activated almost every potential supporter, he might be approaching his electoral ceiling.

Unlike the primary, the general election will be open to about 1.8 million unaffiliated, Republican, and third-party voters. Mamdani’s challengers are likely to have an advantage among most of these non-leftist voters. The Manhattan Institute’s focus group of 11 New Yorkers from across the city’s boroughs and different backgrounds also revealed that “soft” Mamdani supporters have serious reservations about him. The young socialist’s polarizing nature, inexperience, and practical inability to accomplish his agenda represent impediments to further inroads. The focus group suggested that Cuomo has the most to gain at Mamdani’s expense.

The polls also suggest that Cuomo is Mamdani’s most viable challenger, despite his loss in the Democratic primary. The former governor spent part of the summer trying to reinvent his image, admitting that he made mistakes in the primary and vowing to press the case against Mamdani. But with most of the union backing that Cuomo previously received now gone to Mamdani, Cuomo has little campaign support on the ground. His reliance on television and direct mailers notably failed in the primary.

These limitations mean that the most significant potential boost to Cuomo’s campaign lies outside of his control: having Adams and Sliwa drop out. A poll released on Wednesday showed that Cuomo would defeat Mamdani in a head-to-head matchup, 52 to 41 percent.

Meantime, Adams and his campaign have been embroiled in two new corruption-related incidents. On August 20, his longtime campaign operative Winnie Greco allegedly attempted to hand a local journalist a potato-chip bag holding a red envelope with $300 in cash. It then emerged that the Adams campaign supporters had been spotted handing out similar envelopes at Chinese-language media events—a custom of “thank you gifts” ostensibly common in China. Adams denies knowledge of such cash handouts.

The next day, the Adams campaign took a second punch when Ingrid Lewis-Martin, the mayor’s longtime confidant and former chief advisor, was arrested on four bribery-related conspiracy indictments. The once-powerful “Lioness of City Hall” is alleged, among other things, to have received $50,000 in exchange for steering emergency migrant contracts and fast-tracking a permit for a Queens karaoke bar. These new charges come on top of Lewis-Martin’s bribery indictments from last year.

Without the prospect of a comfortable career in law or another lucrative profession, Adams has limited alternatives for life after office. His appeal would diminish further if the corruption-tarred mayor finishes in fourth place and helps deliver the election to Mamdani—an outcome the polls currently predict. Were he to drop out of the race, by contrast, he could plausibly claim to have spurned personal ambition to spare the city from extremism. He then might parlay his one-term mayoralty into a local media role—perhaps as an on-air critic of the next mayor, whoever that might be.

Sliwa has likewise vowed to remain in the race. The Guardian Angels’ founder spoke, sans beret, with Manhattan Institute scholars last week (including me) to explain his vision for New York. He decried bail reform, offering support for President Donald Trump’s recent executive order discouraging the elimination of cash bail and blasting Cuomo for signing bail reform into law in 2019.

Sliwa’s theory of the race hinges on maintaining his 28 percent share of the vote from 2021, plus potentially a few points from his “Protect Animals” ballot line. In a split field with multiple candidates to Mamdani’s right, however, that goal is far from a given.

President Trump remains the race’s biggest X factor. The former New Yorker said in August that he would not let a “communist lunatic destroy New York.” He also intimated that he would send federal law-enforcement officers into the city, which would change the focus of the race from affordability, Mamdani’s strong suit, to two of his weakest areas—personal competence and public safety.

Such a move could also backfire, galvanizing New Yorkers to vote for the furthest-left candidate to spite the president. Should Mamdani win, Trump is certain to attempt to pick up seats in next year’s congressional midterms by stressing the dangers of socialism extended nationwide.

Money is starting to pour into the race, which is likely to break funding records given that general elections are usually uncompetitive. In mid-August, the pro-Cuomo Fix the City PAC raised $1.26 million in a week, while Mamdani’s campaign took in $1.05 million. But the primary election—in which Cuomo and Fix the City outspent Mamdani by a record $22 million to $8 million—demonstrated that money matters less than social media stardom and volunteer boots on the ground.

While the millennial Mamdani is unquestionably the most gifted communicator on the ballot, Adams has demonstrated some acumen in recent posts on Instagram and TikTok. Sliwa has shown his social-media followers videos of him approaching the city’s street homeless and cleaning parks. Cuomo is attempting a twofold strategy—sporting short sleeves in one-on-one talks with small-business owners while also holding suited press conferences seated at a long table and flanked by flags—the latter a format that recalls his popular daily news conferences during the Covid pandemic.

In the most important general election of New York’s modern era, the boldest move for some of Mamdani’s challengers would be to step aside. That’s the toughest request of all for big egos.

Photos: Spencer Platt/Getty Images (left) / Roy Rochlin/Getty Images (center) / Alex Kent/Getty Images (right)

• Who/What/When/Where/Why: Carl Sagan, mid–late 20th century U.S. astronomer and popularizer, sparked debate by widely communicating science to the public and faced professional pushback for doing so.
• Sapolsky's characterization: Robert Sapolsky labeled the phenomenon "Saganized," meaning that heavy public engagement was seen as destroying a scientist's research career.
• Academic setbacks and roles: Sagan was denied tenure at Harvard, became a full professor and laboratory director at Cornell in 1971, and was denied U.S. National Academy of Sciences admission in 1991.
• Perceived reasons for snubs: Critics and some NAS voters viewed popularizers as oversimplifying, self-indulgent, and producing no tangible scientific benefits, interpreting public engagement as a substitute for rigorous research.
• Caplan's point: Arthur Caplan wrote that Sagan's scientific work was discounted not for quality or quantity but because of his association with popular science outreach.
• Empirical counterevidence: Susana Martinez-Conde found most science communicators incur no net career penalty and may gain slightly, though they receive few institutional rewards for outreach.
• Incentive structure and consequences: Academic incentives prioritize journal publication and grant acquisition over public communication, contributing to an insular science culture and political vulnerability amid funding cuts.
• Sagan's legacy: By making science accessible and politically popular, Sagan helped enlarge public and policymaker support for research, producing tangible benefits for the scientific enterprise.

---

Astronomer and science communicator Carl Sagan had – and almost three decades after his untimely death, still has – a great many fans amongst the general public. Sagan's fellow scientists, on the other hand, often frowned upon him. American neuroscientist and primatologist Robert Sapolsky explained why:

Carl Sagan with his billions and billions of stars, he’s like the most successful science writer of his time, and as a result of doing that, he totally destroyed his scientific career. And the snotty term that’s used for it among scientists is, that one gets “Saganized.” There’s a presumption that if you’re spending so much time doing this that you can’t possibly do good, serious science any more.

That Sagan's career was "destroyed" is a bit of an overstatement. Sagan became a full professor at Cornell University in 1971 and served as director of the Laboratory for Planetary Studies and the David Duncan Professor of Astronomy and Space Sciences.

However, before that, he was denied tenure at Harvard. There, colleagues chided that he was too focused on engaging with the public, preferring celebrity to rigorous scientific pursuits. In 1991, more than a decade after the television series Cosmos propelled Sagan to timeless fame, he was denied admission to the U.S. National Academy of Sciences. Sagan's biographers argued that the snub originated from an elitist view that science popularizers like Sagan dangerously oversimplify, are self-indulgent, and produce no tangible benefits. In essence, National Academy of Sciences voters saw Sagan as building his brand in the public sphere because he couldn't actually hack it in the ivory tower.

"His own scientific work was discounted, not due to its quality or quantity, but by dint of association with an author eager and committed to popularizing science and an elitist belief that an unsophisticated public was not worth the time of truly top-flight minds," Arthur Caplan, a professor of bioethics at New York University's Grossman School of Medicine, wrote recently in an op-ed published to EMBO Reports.

Does communicating science to the public still hold back one's career? Spanish-American neuroscientist and science writer Susana Martinez-Conde researched the matter about a decade ago. 

"Most disseminators incur no net penalty in their careers—and may even benefit slightly," she found. "Yet they obtain few or no institutional rewards for their communication activities."

So it doesn't seem like "Saganization" persists today. After all, many scientists are on social media and often speak to the public or news media. Still, at the same time, scientists are not widely incentivized to communicate their research to the public. 

To advance their careers, scientists must publish in scientific journals, not in popular magazines. To procure grant money that fill the coffers of their academic institutions, they must appeal to government bureaucrats, not the general public. This incentive structure maintains science as an insular, elite endeavor, separate from broader society. We're witnessing the effects now.

Populist politicians are slashing science funding because they and their constituents don't believe in its benefits. They might even see scientists themselves as parasites of public dollars.

In this toxic environment, will scientists choose to remain siloed? Will they continue to compete against eachother for slices of an ever-shrinking pie of public research funding? Or will they dismount their high horses and interact with the public, sharing why their research is important? Communicating one's work is not an unreasonable price to pay for receiving public funds.

Historically, scientists may have frowned on people like Carl Sagan, viewing them as subpar researchers. In reality, they need them. By sharing science's insights with laypersons in an easy-to-digest and humble manner, Sagan made science politically and publicly popular. Policymakers in turn enlarged the pie for his fellow researchers, which benefitted both science and society. Sagan's colleagues shouldn't have sneered at him. They should have thanked him.