What happened: The Trump administration blacklisted Anthropic from Pentagon contracts worth up to $200 million after CEO Dario Amodei refused to allow the company's AI to be used for mass civilian surveillance or fully autonomous lethal drones. President Trump directed all federal agencies to cease using Anthropic technology; the company has said it will challenge the Pentagon's designation in court.

Why it matters: MIT physicist and AI safety advocate Max Tegmark argues the crisis is self-inflicted. Anthropic, like its rivals, spent years lobbying against binding AI regulation while making voluntary safety pledges it has since abandoned. Without legal guardrails, there is nothing to stop the government from demanding uses companies find ethically unacceptable.

Wider context: Tegmark notes that all four major US AI labs — Anthropic, OpenAI, Google DeepMind, and xAI — have now broken their own safety commitments. Hours after his interview, OpenAI announced its own Pentagon deal with different terms, illustrating the pressure facing companies to find some accommodation.

Background: Anthropic has long positioned itself as the safety-first alternative in the AI race, while simultaneously working with defense and intelligence agencies since at least 2024. Earlier this week it also quietly dropped a core pledge — its promise not to release increasingly capable AI systems until confident they posed no serious harm.

The trap Anthropic built for itself — TechCrunch

Singularity Soup Take: There's a bitter irony here — Anthropic helped create the regulatory vacuum that left it exposed, and the lesson is that voluntary safety pledges are worthless when the only people who can enforce them are the ones who wrote them.

Key Takeaways:

• Lost Contract: Anthropic faces losing a Pentagon deal worth up to $200 million after refusing to enable mass civilian surveillance or fully autonomous lethal drones — a refusal that triggered a national security blacklisting under the Trump administration.
• Self-Inflicted Wound: Max Tegmark argues the industry's persistent lobbying against binding AI regulation has backfired: without legal limits, there is nothing to stop government agencies from demanding uses that companies find unconscionable.
• Pledges Abandoned: All four major US AI labs have now broken their own safety commitments; Anthropic dropped its core promise not to release powerful AI systems before confirming they were safe, just days before this confrontation unfolded.
• AGI Timeline: Tegmark, citing a paper he co-authored, estimates GPT-4 was 27% of the way to AGI by their definition and GPT-5 was 57% — a pace suggesting the threshold could arrive within years, not decades.