Someone who recently worked inside one of the world’s leading AI companies is now publicly warning that the technology he helped build could eventually threaten humanity.
Jacob Coxon, a former researcher at Anthropic who also previously worked at OpenAI, resigned from Anthropic in September 2026 and accused both companies of racing toward increasingly powerful AI without adequate guarantees that future systems will remain under human control.
His warning was dramatic: Coxon wrote that people building advanced AI “earnestly believe that it could kill us all by the end of the decade.” But that statement is a warning about a possible future scenario—not evidence that AI will destroy humanity by 2030.
So what exactly is Coxon worried about, and how seriously should the public take the risk?
Who Is the Former Anthropic Researcher Sounding the Alarm?
Coxon spent roughly three years working on AI research across OpenAI and Anthropic, focusing on pretraining—the stage where large AI models learn from enormous amounts of data and develop many of their core capabilities.
In an interview with WIRED after leaving Anthropic, Coxon said the pace of AI improvement had become central to his concerns. He pointed to rapid progress in areas including coding, cybersecurity and mathematics and said people inside frontier AI labs increasingly see the next few years as crucial.
That background makes his warning notable. He was not simply observing AI development from outside the industry; he had worked directly on making advanced models more capable.
Why Did He Leave Anthropic?
Coxon said he resigned because he believed the competition among leading AI companies was creating unacceptable risks.
He accused both Anthropic and OpenAI of “racing straight to self-improving superintelligence and gambling with our lives.” According to Axios, he left Anthropic about two months before his equity would have vested, saying afterward that he no longer had a financial incentive to increase Anthropic’s valuation.
Coxon has also acknowledged differences between the companies. He has described Anthropic as taking AI safety seriously, but argues that even safety-conscious companies can face pressure to keep advancing if competitors are doing the same.
Anthropic, for its part, says it has long recognized that increasingly capable AI could bring both major benefits and unprecedented risks. The company maintains a Responsible Scaling Policy designed to evaluate and reduce catastrophic risks as models become more powerful.
Why Is 2030 Being Mentioned?
Coxon did not present 2030 as a scientifically established deadline.
His phrase was “by the end of the decade.” In his WIRED interview, he went even further in emphasizing the near term, saying that people he worked with viewed the next “year or two” as a critical period for deciding whether advanced AI could be developed safely.
The concern centers partly on what happens if AI becomes much better at AI research itself.
Today’s systems can already write code, analyze technical problems and assist researchers. A future system that becomes highly capable at designing algorithms, running experiments or improving other AI systems could potentially accelerate AI development further.
Researchers sometimes describe a more extreme version of this idea as recursive self-improvement: an advanced AI helps create a more capable AI, which then becomes even better at creating the next generation.
Whether such a rapid feedback loop will actually occur—and how soon—is highly uncertain.
What Is the Worst-Case Scenario He Is Worried About?
The concern is not that today’s chatbot suddenly decides to attack humanity.
The more serious hypothetical involves future AI systems that are far more capable, autonomous and able to act in the real world.
A highly capable system might eventually be able to write sophisticated software, conduct cyber operations, manipulate people, control other computer systems or help develop new technologies. If such a system pursued goals that conflicted with human intentions, researchers worry that humans could struggle to stop it.
Possible pathways discussed by AI safety researchers include cyberattacks, interference with critical infrastructure, misuse in biological or military applications and systems attempting to avoid shutdown or oversight.
The 2026 International AI Safety Report makes an important distinction: current AI systems do not yet have the combination of capabilities that researchers believe would be needed for a true loss-of-control scenario. However, researchers have observed early signs of relevant abilities, including greater autonomy, long-term planning and the ability to recognize when models are being evaluated.
Could AI Really Threaten Humanity?
This is where the debate becomes much more complicated.
Coxon is not alone in taking catastrophic AI risk seriously. More than 700 AI researchers and public figures have signed a Center for AI Safety statement saying that reducing the risk of extinction from AI should be treated as a global priority. Signatories include prominent AI researchers and leaders from major technology companies.
After Coxon’s resignation, Anthropic alignment science lead Evan Hubinger publicly agreed that AI could potentially cause human extinction and said his personal estimate was greater than a 10% chance within the next decade.
That number belongs to Hubinger, however—not Coxon—and it should not be treated as a measured scientific probability.
There is also substantial disagreement.
The International AI Safety Report says expert opinions about future loss of control vary widely. Some researchers believe extinction-level outcomes are plausible. Others consider such scenarios implausible because they make different assumptions about future AI capabilities, behavior and deployment.
In other words, there is no scientific consensus that AI will threaten humanity by 2030.
There is also no consensus that the possibility can safely be ignored.
Why This Warning Matters Now
The debate matters because Anthropic, OpenAI, Google DeepMind and other companies are continuing to build models with stronger reasoning, coding and autonomous capabilities.
Safety researchers are therefore asking a practical question: Can safeguards improve as quickly as capabilities do?
Anthropic’s current Responsible Scaling Policy evaluates risks involving areas such as cybersecurity, biological threats, autonomous behavior and automated AI research. The company has also argued that increasingly powerful systems may require stronger external testing and government oversight.
The concern is no longer coming only from researchers who have left AI companies. On September 12, 2026, Reuters reported that Anthropic CEO Dario Amodei called on AI companies to deliberately moderate the pace at which they advance model capabilities. He proposed greater use of independent evaluators, coordination among leading AI developers and international cooperation to create more time to manage emerging risks.
The difficulty is competition. If one company slows down while others continue advancing—or if nations see advanced AI as strategically important—individual companies may have powerful incentives to keep moving.
That is why the AI safety debate increasingly involves not only better technology, but also independent testing, regulation, industry coordination and international cooperation.
Bottom Line
Jacob Coxon’s warning does not prove that AI will destroy humanity by 2030.
What it does reveal is that some people working close to frontier AI development believe the possibility of losing control over much more capable future systems deserves serious attention now.
Current AI systems do not yet have the combination of capabilities that researchers believe would be required for a true loss-of-control scenario. The uncertainty lies in how quickly those capabilities could develop—and whether human safeguards, institutions and regulations can keep pace.
The most important question may therefore be less dramatic than whether AI will literally “kill everyone.”
It is whether humanity can build systems powerful enough to transform science, business, national security and everyday life while still maintaining meaningful control over what those systems do.
Sources
WIRED — The AI Researcher Who Just Quit Anthropic Says It’s ‘Crunch Time for Humanity’
https://www.wired.com/story/anthropic-researcher-quits-jacob-coxon-ai-fears-humanity
Axios — Anthropic Whistleblower Gave Up His Equity to Leave the Company
https://www.axios.com/2026/09/09/anthropic-researcher-ai-warning-interview
Reuters — Anthropic CEO Urges AI Companies to Slow Model Development
https://www.reuters.com/business/anthropic-ceo-urges-ai-companies-slow-model-development-2026-09-12/
Anthropic — Responsible Scaling Policy
https://www.anthropic.com/responsible-scaling-policy
International AI Safety Report 2026
https://internationalaisafetyreport.org/publication/international-ai-safety-report-2026
Center for AI Safety — Statement on AI Extinction Risk
https://safe.ai/work/statement-on-ai-extinction-risk