Lacking comprehensive AI legislation, the US under the Trump administration is increasingly governing frontier AI case by case, through pre-release government reviews and export controls. OpenAI's GPT-5.6 series and Anthropic's Claude Fable 5 and Mythos 5 became the test cases.
June 2, 2026 · Executive Order on AI
America's "Shadow AI Policy": Governing Frontier Models One Case at a Time
With no comprehensive federal AI law, the US now shapes model releases through pre-release government reviews and export controls. OpenAI's GPT-5.6 and Anthropic's Claude Fable 5 & Mythos 5 became the first test cases of this "voluntary" — but decisive — gatekeeping.
12 days
GPT-5.6 held from limited preview to public release, awaiting a White House review
30 days
Maximum early-access review window granted to the government before release
<1%
Of government staff hold AI PhDs — a core capacity gap
Funding vs. the Mission
CAISI, the body tasked with AI standards, operates on roughly one-sixth of what executing the AI Action Plan is estimated to cost.
$15M
CAISI operating budget
$84M/yr
Est. AI Action Plan cost
Case 1 · Light touch
OpenAI GPT-5.6 (Sol · Terra · Luna)
Late June: limited preview for trusted partners at government request → early July: broad release after review. A 12-day gate.
Case 2 · Heavy hand
Anthropic Claude Fable 5 & Mythos 5
Disabled after release under export controls — a full halt for foreign nationals. Restored weeks later once guardrails & classifiers blocking cyber tasks were added.
The de facto release pipeline
Developer builds
→
Early gov. access
→
Guardrails &
→
Release, once gov.
Nominally voluntary — yet in practice government satisfaction has become a precondition for release.
Supportive
Sam Altman (OpenAI): a "productive process," promising improvement — noting the world would grow uncomfortable if firms lacked confidence in safety claims.
Critical
Asad Ramzanali: neglects privacy, antitrust & transparency.Kevin Frazier (Cato): too few technical experts; lean on states.Rep. Josh Gottheimer: "far too much confusion" in the vetting process.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…