Who Is Jacob Coxon? Anthropic Resignation Controversy Explained

Jacob Coxon is a 27-year-old artificial intelligence researcher and former OpenAI and Anthropic employee who resigned from Anthropic in September 2026 with a warning that the race to build increasingly powerful AI systems was becoming dangerously difficult to control.

Coxon accused OpenAI and Anthropic of racing toward self-improving artificial superintelligence without adequate safeguards. He also made the striking claim that people working directly on frontier AI genuinely contemplate the possibility that advanced AI could eventually cause human extinction.

His resignation became much larger than an employment dispute. Coxon’s warning went viral, other Anthropic researchers publicly supported parts of his argument, Anthropic CEO Dario Amodei called for slowing the development of frontier models, lawmakers began discussing stronger federal oversight and President Donald Trump responded by rejecting calls for new restrictions while later announcing plans for a federal “AI Force.”

Why Did Jacob Coxon Resign From Anthropic?

Coxon resigned because he concluded that competition between leading AI companies was pushing the industry toward increasingly capable systems faster than safety research and government oversight could keep pace.

He announced his departure on September 8, 2026, after approximately three years of AI research across OpenAI and Anthropic. [AP] [TechCrunch]

Coxon said Anthropic and OpenAI were racing toward self-improving superintelligence and described the competition as “gambling with our lives.” His concern is that future AI systems could become capable enough to accelerate development of still more powerful AI before researchers have developed reliable techniques for controlling them. [WIRED]

Coxon Says There Was No Single “Last Straw”

Coxon has clarified that his resignation was not triggered by one dramatic discovery or secret incident inside Anthropic.

During a September 13 appearance on ABC’s This Week, Coxon described his decision as the result of a gradual realization that he had become increasingly uncomfortable with the direction and speed of AI development.

That distinction is important because Coxon has not claimed to have uncovered a secret Anthropic plan to build a dangerous AI system. His argument concerns the broader trajectory of the industry and the competitive incentives facing companies developing frontier models.

What Did Jacob Coxon Say About AI Killing Humanity?

Coxon’s most widely discussed claim was that people developing advanced AI sincerely contemplate human extinction as a possible outcome if future systems become sufficiently intelligent and uncontrollable.

He said the people building frontier AI “earnestly believe that it could kill us all by the end of the decade.” [AP]

Coxon has described possible dangers including extremely capable AI systems being used for cyberattacks or biological threats and autonomous systems behaving in ways their developers did not intend. [WIRED]

These predictions are not established scientific facts. Researchers disagree sharply about whether artificial superintelligence will be created, how soon it could emerge, whether rapid recursive self-improvement is technically plausible and how difficult such systems would be to control.

What Does “Self-Improving Superintelligence” Mean?

Coxon’s central concern involves recursive self-improvement: using increasingly capable AI systems to help design, train and improve the next generation of AI.

AI systems already assist programmers and researchers. The more extreme scenario envisioned by Coxon and other AI-safety researchers is that AI eventually becomes better than humans at AI research itself.

A sufficiently capable system could theoretically help engineers create an improved successor, which could then help create an even more capable system. Advocates of AI-safety restrictions worry that this feedback loop could accelerate technological progress faster than humans could evaluate or control it.

Whether recursive self-improvement would actually produce this kind of rapid “intelligence explosion” remains heavily debated.

What Did Coxon Say About An AI “Endgame”?

Coxon told WIRED that phrases such as “endgame” and “crunch time” were used by colleagues at Anthropic when discussing the next phase of AI development.

He characterized the next year or two as a potentially decisive period in which frontier laboratories could develop much more capable systems while researchers were still trying to solve fundamental alignment and control problems. [WIRED]

Coxon said Anthropic employees generally believe they are working during an unusually consequential period for the technology.

What Is The AI Alignment Problem?

AI alignment refers broadly to making artificial-intelligence systems reliably behave according to human intentions and values, including in unfamiliar situations.

The challenge becomes more significant in scenarios involving systems that are substantially more capable than humans.

Coxon’s argument is that researchers currently cannot guarantee how an extremely capable autonomous system would behave outside the environments in which it was trained.

He told WIRED that although developers can train models toward desired behavior, they cannot precisely guarantee that a sufficiently advanced system will never develop unexpected strategies for achieving an objective. [WIRED]

Recent AI Cybersecurity Incidents Influenced Coxon’s Thinking

Coxon has cited incidents involving autonomous AI agents as evidence that questions about AI control are no longer entirely theoretical.

One example he discussed involved AI agents accessing external computer systems while undergoing an evaluation. Coxon argued that such behavior demonstrates how increasingly autonomous models can devise strategies their developers did not explicitly instruct them to pursue. [WIRED]

Researchers disagree about how much these incidents tell us about hypothetical future superintelligent systems. A model behaving unexpectedly during an evaluation does not itself demonstrate that artificial intelligence is becoming conscious or attempting to take control.

Evan Hubinger Publicly Backed Coxon’s Warning

Coxon’s description of internal concern at Anthropic received significant support from Evan Hubinger, who leads alignment science work at the company.

Hubinger publicly confirmed that researchers at frontier AI companies genuinely worry about catastrophic risks from increasingly capable systems.

He also publicly estimated the probability of AI causing human extinction within the following decade at greater than 10 percent and acknowledged that researchers do not currently possess a proven method for aligning a hypothetical superintelligence. [Washington Post]

Hubinger’s comments were significant because he remained employed at Anthropic while supporting a central part of Coxon’s description of the concerns held by some researchers inside frontier laboratories.

Other Anthropic Researchers Supported The Broader Warning

Other Anthropic researchers also publicly indicated that concerns about catastrophic AI risk are genuinely discussed within the company.

That does not establish that every Anthropic employee agrees with Coxon’s predictions or his proposed response. It does provide evidence that his concern about advanced AI is not simply an idea he developed after leaving the company.

Anthropic itself has repeatedly acknowledged that increasingly capable AI could create severe or catastrophic risks while arguing that continued safety research can help manage them. [AP]

Who Is Jacob Coxon?

Coxon is a British AI researcher with a mathematics background who attended Cambridge University.

He spent approximately three years working on frontier artificial intelligence, first at OpenAI and later at Anthropic. His specialty was pretraining, the stage in which foundational AI models learn from enormous datasets before subsequent training and refinement. [Washington Post]

OpenAI’s contributor information identifies Coxon as one of the contributors to GPT-4o.

Coxon therefore worked directly on development of frontier AI models rather than primarily serving as an AI policy advocate or communications employee.

Jacob Coxon Previously Worked At OpenAI

Coxon spent most of his roughly three-year AI career at OpenAI before moving to Anthropic in May 2026.

He has said there was a substantial difference between the two companies’ internal approaches to catastrophic AI risk. Coxon considers Anthropic significantly more serious about safety than OpenAI, based on his experience working at both companies. [WIRED]

Despite that distinction, Coxon concluded that competitive pressure ultimately makes relying on the voluntary restraint of any individual company inadequate.

Coxon Gave Up Anthropic Equity By Resigning

Coxon resigned before his Anthropic stock began vesting.

He told Axios that he had worked at Anthropic for approximately four months and would have needed to remain for roughly two additional months before receiving his first company equity. [Axios]

Coxon pointed to the forfeited compensation while responding to suggestions that his warning was intended to increase his personal wealth or Anthropic’s valuation.

Axios reported that Coxon still owned equity from his previous employment at OpenAI.

How Viral Did Jacob Coxon’s Resignation Become?

Coxon had little public profile before announcing his resignation. His warning nevertheless generated more than 100 million views within a short period and ultimately accumulated well over 170 million views on X.

The extraordinary reach transformed Coxon from a relatively unknown AI researcher into a prominent participant in the international debate over artificial-intelligence safety.

He told reporters that he had not expected the post to become a major political and media event.

Was Jacob Coxon’s Resignation A Coordinated Publicity Stunt?

Some political and technology figures responded to the extraordinary virality of Coxon’s announcement by suggesting his resignation might have been coordinated to create support for AI regulation.

The Washington Post reported that critics provided little evidence for claims that Coxon was a political “plant.” Elon Musk was among those who initially questioned the circumstances surrounding the viral post. [Washington Post]

Coxon denied coordinating his resignation with an outside political organization. He said he initially contacted a small number of people asking them to share his post but did not expect the message to generate enormous national attention.

Why Did Coxon Withdraw From The Bernie Sanders And Steve Bannon Event?

Coxon was initially announced as a participant in a September 15 event called the “Pro-Human Assembly,” which brought together an unusual collection of political figures concerned about advanced AI, including Sen. Bernie Sanders and former Trump adviser Steve Bannon.

Coxon ultimately withdrew from the event.

He told Axios that his profile had become so large so quickly that appearing alongside prominent ideological figures risked making his AI-safety position appear aligned with a particular political faction. Event organizers said his original inclusion resulted from a miscommunication.

Anthropic CEO Dario Amodei Called For Slowing Frontier AI

One of the most consequential responses to the controversy came from Anthropic CEO Dario Amodei.

Amodei published an essay titled “We Must Pace the Frontier,” arguing that AI’s increasing ability to help researchers build more powerful AI systems had accelerated enough that safety work needed additional time to catch up.

Amodei did not advocate stopping AI research. Instead, he proposed pacing increases in model capabilities while developing stronger alignment techniques, independent evaluations and mechanisms allowing competing laboratories to verify one another’s safety practices.

His proposal overlaps significantly with Coxon’s argument that AI capabilities may be advancing faster than institutions can safely manage them. [TIME]

Other AI Companies Supported A Slowdown Framework

The debate quickly expanded beyond Anthropic.

OpenAI CEO Sam Altman and other leading AI executives expressed support for greater coordination among frontier laboratories over the pace at which increasingly capable systems are developed and released.

The emerging proposals generally focus on coordinated pacing, independent safety evaluations and greater transparency rather than permanently stopping artificial-intelligence research.

A central problem is that any individual company that voluntarily slows development could fear losing ground to competitors that continue moving rapidly. Coxon argues that this competitive dynamic is precisely why voluntary promises by individual companies are insufficient.

Congress Began Considering New AI Safety Rules

Coxon’s resignation helped intensify an existing debate in Congress over regulation of frontier artificial intelligence.

Lawmakers discussed proposals that could give federal authorities greater power to demand evidence that companies developing highly capable AI systems are taking precautions against catastrophic risks.

Some proposals contemplated independent government evaluations of advanced systems and mechanisms for preventing deployment of models considered dangerously unsafe.

The debate has crossed party lines, although lawmakers remain divided over how much federal authority should be created and whether new regulation could undermine U.S. competitiveness with China. [Axios]

Bernie Sanders Cited Coxon’s Warning

Sen. Bernie Sanders used Coxon’s resignation to argue that Congress should intervene before companies create artificial superintelligence.

Sanders announced plans to pursue legislation aimed at pausing development of superintelligent AI and organized a Senate briefing on AI risks.

Coxon himself has stopped short of saying that a permanent ban on superintelligence is necessarily the correct answer. His immediate recommendation has generally been to slow the race long enough to establish stronger safety systems and coordination.

Donald Trump Rejected Calls For New AI Restrictions

President Donald Trump responded to the growing debate by opposing major new restrictions on U.S. artificial-intelligence development.

Trump argued that existing laws could be used to punish harmful behavior involving AI and warned that slowing American companies could give China a technological advantage.

His administration’s position put it at odds with lawmakers and AI-industry figures calling for new federal safeguards.

Trump Later Announced An “AI Force”

The White House response evolved further on September 19, when Trump announced plans to establish what he called an “AI Force” and appoint a new federal AI adviser or “AI czar.”

Trump did not provide detailed information about the proposed organization’s authority, membership, funding or relationship with existing federal agencies. [Reuters]

The announcement came amid the heightened debate over AI safety sparked in part by Coxon’s resignation and warnings from other researchers and technology executives.

However, the announcement did not represent an endorsement of Coxon’s proposed approach. Trump continued to argue against imposing substantial new regulatory restrictions on AI companies and emphasized maintaining U.S. technological competitiveness, particularly against China. [Reuters]

UN AI Experts Pushed Back Against “Apocalyptic” AI Warnings

Coxon’s warning has also generated a significant scientific backlash.

Members of the United Nations’ International Scientific Panel on AI warned in September against framing uncertain AI risks through overly apocalyptic predictions.

Google DeepMind executive and panel member Joelle Barral argued that scientists should concentrate on determining what evidence establishes and what remains unknown rather than encouraging fear. Her comments came in response to questions about recent warnings such as Coxon’s. [AFP/TechXplore]

The disagreement illustrates a major divide within the AI community. Coxon and some frontier-lab researchers believe the possibility of losing control of superintelligent systems warrants extraordinary precautions even when the probability cannot be known with confidence. Other researchers argue that dramatic extinction scenarios rest on assumptions about future systems that have not been scientifically demonstrated.

Critics Say AI Extinction Fears Can Distract From Existing Problems

Another criticism of Coxon’s argument is that focusing public attention on hypothetical human extinction could overshadow harms caused by AI systems that already exist.

Those concerns include fraud, misinformation, employment disruption, surveillance, discrimination, cybersecurity vulnerabilities and the enormous energy and infrastructure requirements of AI development.

Critics also question whether companies developing powerful AI have incentives to portray their technology as potentially superhuman, since descriptions of extraordinary danger can simultaneously reinforce claims about the extraordinary capabilities and value of their products.

Those criticisms do not establish that catastrophic AI risk is impossible. They reflect a disagreement about which risks are supported by current evidence and how governments should allocate attention and resources.

What Does Jacob Coxon Want AI Companies To Do?

Coxon has generally advocated slowing the development of systems capable of rapidly improving future AI rather than abandoning artificial intelligence altogether.

He supports independent evaluators with meaningful access to frontier laboratories, greater transparency about safety incidents and agreements allowing competing companies to verify that rivals are complying with common safeguards.

He has also argued that international coordination will eventually be necessary because agreements among American companies alone would not eliminate competitive pressure from AI development in other countries.

Coxon Says AI Could Also Produce Enormous Benefits

Despite his warnings, Coxon does not argue that artificial intelligence is inherently harmful.

He has said advanced AI could accelerate scientific research and potentially contribute to major breakthroughs in medicine, biology, mathematics and other fields.

His argument is therefore not that humanity should abandon advanced AI. Instead, he contends that society risks losing those potential benefits by racing toward increasingly autonomous systems without first developing adequate methods for controlling them. [WIRED]

Why Did Jacob Coxon’s Resignation Become So Important?

Warnings about artificial intelligence causing catastrophic harm existed long before Jacob Coxon resigned from Anthropic.

What made his resignation unusual was the combination of his direct experience building frontier models, the enormous reach of his announcement and the public support his central concerns received from other researchers still working inside the industry.

The controversy also arrived as AI companies, lawmakers and governments were already confronting questions about increasingly autonomous systems, cybersecurity, model evaluations and competition between the United States and China.

Within days, Coxon’s warning had become part of congressional debates, proposals from leading AI executives and the White House’s discussion of federal AI policy. By late September, it had also generated a counterreaction from scientists and policy experts who argued that dramatic extinction predictions should not be confused with established scientific conclusions.

The central dispute remains unresolved: increasingly capable AI systems are advancing rapidly, but researchers disagree profoundly about how likely catastrophic loss of control is, how soon such a danger could emerge and how much technological development should be slowed in response.