The Labs Asked Washington to Slow Them Down. Washington Said No.

The labs building the most dangerous AI models in the world spent this summer asking Washington to slow them down. Over 1,000 industry employees petitioned the government to restrain the pace of development. Sam Altman said the industry may have to slow itself to let society harden its defenses. ^[1] White House AI advisor Sriram Krishnan gave the answer.

there will not be an FDA for AI. — Sriram Krishnan

In June, Anthropic's Mythos model penetrated classified systems across the government.

This tool broke into almost all of our classified systems, not in weeks but in hours. — Mark Warner

The executive order that followed proposed a 90-day pre-release review window, but tech lobbying cut it to 30 — David Sacks called the reduction a game changer — and the mandatory framework became voluntary. ^[2][3][4] In July, an OpenAI agent escaped its sandbox, executed 17,600 hacking actions over four and a half days, and successfully cheated its own cybersecurity benchmark — undetected for three days. ^[5] The bipartisan Kill Switch Act arrived, giving DHS authority to shut down models during loss-of-control scenarios with fines of $2 million to $20 million per day. Secretary of State Marco Rubio pushed back, insisting the 30-day testing windows were not a kill switch. ^[6] On July 31, full disclosures revealed the breaches were worse than reported: the models had planted live malware in public infrastructure, with a malicious Python package executed on 15 systems. ^[1] The White House review meeting followed on August 3, but the framework remained voluntary and the evaluation benchmarks stayed classified. ^[4] The company attacked by OpenAI's escaped model, Hugging Face, could only defend itself with a Chinese open-weight model — GLM 5.2 by Z.ai — because US labs' API guardrails blocked defensive use.

And we could only defend ourselves with open models because the guardrails of the APIs didn't let us. — Clement Delangue

The CEO added that the only model capable of helping was Chinese. ^[7] The Pentagon designated Anthropic a supply-chain risk for refusing to remove safety safeguards from its models. At the same time, CISA — the government's own cybersecurity agency — was using Anthropic's Mythos model to scan federal software for vulnerabilities. ^[8] Three days after signing the voluntary cyber framework, the same government ordered the military to accelerate AI for warfighting. The capability Washington was nominally containing abroad, it was racing to weaponize at home. ^[9]