Skip to main content
AI ModelsEvidence

How the US actually governs frontier AI, and how lightly

US frontier-AI governance runs on two tracks: binding executive action and voluntary industry commitment.

The August 2026 model-testing framework sits on the second, it grants the government up to 30 days of pre-release access to frontier models, but it is explicitly not a licensing regime, so no lab faces a pre-approval gate.

US Frontier-AI Policy Tracker

Data date
August 2026
Source class
Reputable secondary
Data points
4
Sources
3
Value basis
4 actual

Caveat: 2023–2025 rows are not seeded, a defensible count needs an agreed inclusion rule for agency guidance, and that rule is not settled.

This page can answer

Two series: federal AI governance actions by year and type, and the frontier labs inside the August 2026 voluntary testing framework.

It cannot answer

  • Compliance or outcomes, the index records participation only.
  • A 2023–2025 action count; the inclusion rule is unsettled so those rows are omitted rather than guessed.
  • State-level or non-US regimes.
Jump to chaptersSource ledger
Evidence chapters

Chapter 01 · Question answered

How many federal AI governance actions, split binding vs voluntary.

US frontier-AI policy actions, 2023 → 2026

Policy actions by year

Policy actions by year

Count of binding and voluntary federal AI governance actions per year.

View data & sources →

Policy actions by year

Rendering the interactive chart…

Expand data table
Download CSV
US frontier-AI policy actions, 2023 → 2026: data table
SeriesValueValue basisSource
Executive orders / directives1ActualCNBC: White House, AI companies agree voluntary framework
Voluntary frameworks1ActualCNBC: White House, AI companies agree voluntary framework

US frontier-AI policy actions, 2023 → 2026: full data table

SOURCECNBC: White House, AI companies agree voluntary framework
VALUE BASISActual
LAST VERIFIEDSep 10, 2026

Caveat: 2023–2025 rows are not seeded, a defensible count needs an agreed inclusion rule for agency guidance, and that rule is not settled.

How this is calculated

Chapter 02 · Question answered

Which frontier labs are inside the framework.

Labs participating in the federal testing framework

Lab participation

Lab participation

Frontier labs that attended the White House session on 4 August 2026.

View data & sources →

Lab participation

Rendering the interactive chart…

Expand data table
Download CSV
Labs participating in the federal testing framework: data table
DimensionSeriesValueValue basisSource
OpenAIParticipating (1 = yes)1ActualBloomberg: OpenAI, Anthropic, Google to join White House AI safety meeting
AnthropicParticipating (1 = yes)1ActualBloomberg: OpenAI, Anthropic, Google to join White House AI safety meeting
GoogleParticipating (1 = yes)1ActualBloomberg: OpenAI, Anthropic, Google to join White House AI safety meeting

Labs participating in the federal testing framework: full data table

SOURCEBloomberg: OpenAI, Anthropic, Google to join White House AI safety meeting
VALUE BASISActual
LAST VERIFIEDSep 10, 2026
How this is calculated
All data for this index

Data table

Download CSV
US Frontier-AI Policy Tracker: full data table
YearSeriesSourceValue basisExecutive actionsVoluntary frameworksLabParticipating
2026policy_actionsCNBC: White House, AI companies agree voluntary frameworkActual11Not publishedNot published
Not publishedlab_participationBloomberg: OpenAI, Anthropic, Google to join White House AI safety meetingActualNot publishedNot publishedOpenAI1
Not publishedlab_participationBloomberg: OpenAI, Anthropic, Google to join White House AI safety meetingActualNot publishedNot publishedAnthropic1
Not publishedlab_participationBloomberg: OpenAI, Anthropic, Google to join White House AI safety meetingActualNot publishedNot publishedGoogle1

Method

How this evidence was built

What it measures

Counts federal AI governance actions by year, split into binding executive action and voluntary industry framework. The August 2026 framework grants the government early access to frontier models for up to 30 days before release and is explicitly not a licensing regime.

Key numbers

  • 2026 carries one binding executive action (June AI-cybersecurity eo) and one voluntary framework (cnbc-wh-framework).
  • OpenAI, Anthropic and Google were the labs at the 4 August 2026 White House session (bloomberg-wh-meeting).
  • The framework grants up to 30 days of pre-release model access and is explicitly not a licensing regime (cnbc-wh-framework).

Caveats

Caveat: the 2023–2025 rows are not yet seeded. A defensible count requires an agreed inclusion rule (does an agency guidance document count as an "action"?) and that rule is not settled, so per the binding rules those tuples are omitted rather than guessed. Only the 2026 row, where the two actions are individually dated and named, is seeded.

Caveat: 2023–2025 rows are not seeded, a defensible count needs an agreed inclusion rule for agency guidance, and that rule is not settled.

Do not use this to

  • Compliance or outcomes, the index records participation only.
  • A 2023–2025 action count; the inclusion rule is unsettled so those rows are omitted rather than guessed.
  • State-level or non-US regimes.

Scope

August 2026

Sources

Last verified 10 Sept 2026

Methods that use this evidence

Use one of these copy-paste methods when you want an AI to work from this evidence.

How relevant was this information?