More Anthropic researchers warn of AI dangers as Musk calls it a 'psyop'
Several Anthropic staff members have publicly warned that the AI technology they build could pose extinction-level risks within a decade, prompting Elon Musk and others to dismiss the concerns as a coordinated 'psyop'.
A day after former Anthropic researcher Jacob Coxon announced his resignation, warning that neither Anthropic nor rival OpenAI were developing AI responsibly, several current employees at the company publicly echoed his concerns.
Coxon said on Wednesday that both companies were "gambling with our lives," and claimed many colleagues privately share his worries but have stayed silent. His post quickly went viral, prompting other Anthropic staff to speak up.
Anna Wang, who works on Artificial General Intelligence Safety at Anthropic and previously worked at Google DeepMind, said many people inside the company want development slowed down until a credible plan exists to manage the risks. She noted there is still no viable scientific plan for handling risks from AI systems capable of recursively improving themselves.
Drake Thomas, another employee, said he respected Coxon's decision, adding that development is moving far too fast to build the level of assurance needed before reaching artificial superintelligence.
Musk suggests a coordinated campaign
Elon Musk responded on X by calling the wave of concern a "setup." Coxon replied with a selfie, insisting his beliefs were genuine. Musk and several conservative commentators, including researcher Parker Thayer, suggested without clear evidence that the episode could be a coordinated effort to build support for AI regulation. Billionaire investor Bill Ackman also drew attention to the theory.
Other Anthropic staff, including safety researcher Samuel Marks and alignment lead Evan Hubinger, said they genuinely believe AI could pose an extinction-level threat, estimating the probability at more than 10% within the next decade.
Not everyone agrees the danger is existential. AI scientist Gary Marcus argued the more pressing risks are nearer-term — AI-generated pathogens, disinformation-fueled conflict, and attacks on critical infrastructure. The same day, Anthropic published a report describing how it had disrupted an attempt to use its AI models to help develop a biological weapon.
