Which frontier labs are inside the framework.
Labs participating in the federal testing framework
Takeaway: Attendance, not compliance: participation is voluntary.
Lab participation
Frontier labs that attended the White House session on 4 August 2026.
Takeaway: Attendance, not compliance: participation is voluntary.
Lab participation
Attendance, not compliance: participation is voluntary.
Rendering the interactive chart…
Expand data table
| Dimension | Series | Value | Value basis | Source |
|---|---|---|---|---|
| OpenAI | Participating (1 = yes) | 1 | Actual | Bloomberg: OpenAI, Anthropic, Google to join White House AI safety meeting |
| Anthropic | Participating (1 = yes) | 1 | Actual | Bloomberg: OpenAI, Anthropic, Google to join White House AI safety meeting |
| Participating (1 = yes) | 1 | Actual | Bloomberg: OpenAI, Anthropic, Google to join White House AI safety meeting |
Data table
Download CSV| Lab | Series | Source | Value basis | Participating |
|---|---|---|---|---|
| OpenAI | lab_participation | Bloomberg: OpenAI, Anthropic, Google to join White House AI safety meeting | Actual | 1 |
| Anthropic | lab_participation | Bloomberg: OpenAI, Anthropic, Google to join White House AI safety meeting | Actual | 1 |
| lab_participation | Bloomberg: OpenAI, Anthropic, Google to join White House AI safety meeting | Actual | 1 |
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