Trump Administration Nears Standards Deal With Big AI

According to the Financial Times, cited by Gizmodo, the Trump Administration and several major U.S. frontier AI companies are expected to announce standards for frontier AI models, particularly around cybersecurity capabilities, within days. The report relies on anonymous people familiar with the talks, who said CAISI and the NSA will be central to the standards once formalized. Reporting names Anthropic, OpenAI, Amazon, Microsoft, and Google as likely parties.

This builds on a rapid escalation in the administration's AI posture. On June 2, 2026, Trump signed an executive order creating a voluntary process letting the government and critical-infrastructure operators request up to 30 days of pre-release access to powerful new AI models, with NIST, DHS, Treasury and the White House given 60 days to define which models qualify. The order followed alarm in Washington over Anthropic's Mythos model, which can identify and exploit software vulnerabilities far faster than defenders can patch them. On Friday, June 12, 2026, the government directed Anthropic to immediately block foreign nationals, including its own employees, from accessing its newer Fable 5 and Mythos 5 models; Anthropic responded by suspending access entirely, including domestically. The directive reportedly followed Amazon CEO Andy Jassy warning officials that Fable 5's anti-hacking guardrails could be bypassed, after Amazon researchers demonstrated a narrow jailbreak. Anthropic disputed the severity, saying the flaws found were minor and discoverable by other publicly available models. By June 15, 76 CISOs, researchers, and executives had signed a public letter to Commerce Secretary Howard Lutnick and National Cyber Director Sean Cairncross calling the ban "dangerous" and warning it hands defenders' best tools to adversaries while China's models remain only months behind. OpenAI has separately agreed to voluntarily limit release of a newer model at the government's request.

Since the benchmarking process is expected to be classified, the public specifics of what standards models must meet may stay opaque, but shared practices across signatory companies should make emerging norms partially inferable. Teams building or deploying frontier-scale models should watch for any CAISI-published guidance or attestation process, whether their models could fall under a "covered frontier model" threshold, and contingency plans in case of a sudden access suspension similar to Anthropic's June episode.

This specific standards-deal report remains preliminary and anonymously sourced, not yet confirmed by the White House, CAISI, or any named company. Key open questions are which companies actually sign on, especially whether Meta joins, whether the announcement slips past the reported timeline, and whether the underlying Anthropic export-control dispute, which Anthropic is contesting, gets resolved or folded into the new standards framework.