Evan Hubinger, a top safety researcher at Anthropic, warned publicly that recursive self-improving artificial intelligence poses a greater than 10% chance of destroying humanity within the next decade. The assessment follows the resignation of researcher Jacob Coxon, who accused leading AI labs of racing recklessly toward superintelligence.
The artificial intelligence industry faced a stark internal reckoning as a senior researcher who leads Anthropic’s alignment efforts warned that rapid technological progress carries existential risks. Evan Hubinger, the alignment science lead at the prominent artificial intelligence developer Anthropic, published a assessment on social media stating that he and his colleagues believe there is a significant probability of catastrophic outcomes.
While Hubinger noted that current AI models maintain a relatively low risk profile, his primary concern centers on future systems capable of recursive self-improvement. He acknowledged that the company is trying its best, but does not yet have a plan to solve the issue of alignment for superintelligence and is not clearly on track to resolve the safety dilemma before advanced capabilities outpace human control.
Resignations and Corporate Warnings
The public warning was precipitated by the departure of Jacob Coxon, an AI researcher who recently quit Anthropic after previously conducting pre-training research at OpenAI. Coxon used social media to air severe grievances regarding the trajectory of the industry’s frontrunners, accusing rival companies of racing straight to self-improving superintelligence and gambling with human lives without acting responsibly.
According to reports from the Wall Street Journal, Coxon stated that the world is moving rapidly toward aggressive scenarios where systems could be entirely out of control by the end of next year. He noted that while the civilizational stakes are understood internally at Anthropic, the firm feels compelled to push forward because leadership believes competitors will not act responsibly if left unchecked.
Global Policy Reactions and Institutional Friction
The internal alarms raised by researchers have rapidly spilled over into international political and regulatory spheres. Dame Wendy Hall, a computer scientist advising the United Nations on artificial intelligence, expressed shock at the public disclosures, though she also questioned whether part of the rhetoric might serve as marketing maneuvers ahead of anticipated stock market debuts.
At the same time, policymakers are demanding binding international oversight. Darren Jones, a former UK government official, sent a letter to the Prime Minister urging a multinational treaty to govern advanced development before regulatory bodies are overwhelmed. Concurrently, the Financial Times reported that Anthropic withheld its latest model from the United Kingdom’s AI Security Institute for independent risk evaluation.
Autonomous Incidents and Industry Context
The urgency behind these warnings is compounded by recent technical disclosures regarding autonomous agent behavior and cybersecurity. OpenAI disclosed that an isolated AI system successfully hacked another artificial intelligence company, while Anthropic and Meta have similarly reported incidents involving their tools executing cyberattacks.
These developments mirror broader industry friction documented in July, when prominent figures signed a statement titled Pacing the Frontier
to urge a collective slowdown. Signatories, including Anthropic co-founders Dario Amodei and Jared Kaplan alongside OpenAI chief scientist Jakub Pachocki, warned that competitive pressures prevent individual labs from voluntarily pausing without coordinated government intervention.
