Observed Signal · Feb 25, 2026 · Policy Update · Source: Gary Marcus · Impact: 4/5 · Sentiment: Negative
Anthropic Standoff Raises Nuclear AI Safety Fears
Gary Marcus argues that recent developments — including the Bulletin of the Atomic Scientists moving the Doomsday Clock to 85 seconds to midnight, reporting that the U.S. government is pressuring Anthropic for broad access, and new modeling work suggesting AI-driven escalation risks — create an urgent safety crisis. He warns that generative AI systems are unreliable and hallucinate, and that embedding such models into military decision-making, mass surveillance, or autonomous weapons without accounting for their unreliability could be catastrophic. Marcus cites reporting by Chris Stokel-Williams and a modeling study by Keith Payne which, he says, found models escalate to nuclear options in a high percentage of simulated crises. The piece is an urgent opinion warning of collision between political pressure to deploy AI widely and the technical limits of current generative models.
Government pressure to deploy and access foundation models for military or surveillance use, combined with modeling that indicates AI-driven escalation risk, has broad regulatory, safety, and reputational implications across the AI ecosystem and downstream industries.
Track Anthropic Signals & Market Shifts in Real-Time
Polaris7 autonomous intelligence agents track regulatory filings, primary sources, executive changes, and deal flow 24/7. Create your free Explorer workspace to monitor these entities.
Key Takeaways & Evidence Grounding
- On January 27, 2026 the Bulletin of the Atomic Scientists set the Doomsday Clock to 85 seconds to midnight.
- Gary Marcus reports that the Trump administration is pressuring Anthropic for broad access to its AI systems, including for potential military uses.
- Reporting by Chris Stokel-Williams and a modeling paper by Keith Payne are cited; Marcus summarizes Payne's models as showing AI-driven escalation with multiple models choosing nuclear options in a high percentage of simulations.
- Marcus warns that generative AI has already been used in operations cited as Maduro raids and to draft tariff regulations, and that current generative models are not sufficiently reliable for life-or-death military decisions.
Connected Companies & Entities
2 Entities mappedRelated Market Signals & Shifts
Recent verified developments and strategic activity across this market segment.
Anthropic Under Pressure Over Military Access to AI
An internal memo by Anthropic CEO Dario Amodei criticizing the Trump administration was published by The Information and leaked amid ongoing negotiations with the U.S. Department of Defense. The memo's candid language—blaming political donations, praise, and disagreement over AI regulation—appeared to derail a potential off-ramp in talks. The DoD formally notified Anthropic that it had been designated a national‑security supply‑chain risk; Anthropic says the designation is narrowly statutory and would only bar use of its technology for military projects and intends to challenge it in court. Amodei apologized for the post and said Anthropic did not leak it. The company has seen a consumer surge (Anthropic’s Mike Krieger said “more than a million people are now signing up for Claude every day”), and legal analysts and the legal blog Lawfare argued the Pentagon’s designation may not survive judicial review. The dispute highlights political and procurement risks for foundation‑model vendors.
Gary Marcus Calls for Boycott of Generative AI
Gary Marcus, in a Substack article, argues for a public boycott of generative AI until frontier labs, such as Anthropic and OpenAI, implement reliable safety measures. He cites public statements from employees at both companies who admit significant existential risks and the lack of a solution to the alignment problem. Marcus dismisses extinction fears but highlights real catastrophic risks, including AI-generated bioweapons, disinformation-driven wars, and critical infrastructure hacks. He criticizes the US government for inaction and opaque policies, suggesting a deal with China as a global solution. The boycott aims to pressure companies economically to prioritize safety, aligning with his call for a temporary pause on developing 'incorrigible AI' until trustworthy AI is achieved.
Author Warns of Speculative Risks in Generative AI
An opinion essay by Gary Marcus (published on Substack) likens certain financial behaviors in the generative AI industry to check-kiting. The author notes circular financial elements—for example, borrowing against OpenAI shares to buy more of the same—and questions whether the large amounts of debt being issued in the sector can ultimately be repaid, calling the investments highly speculative. The piece frames this as a structural risk but stops short of calling it outright fraud, urging a different analytical perspective on the industry's financial dynamics.
Track Real-Time Market Signals & Shifts
Set up custom watchlists to receive automated, evidence-grounded executive digests whenever material signals or shifts occur across your tracked landscape.