The AI testing process Anthropic has advocated at CAISI and the White House involves more rigorous tests for frontier models than for off-frontier models.

Judge as of 2026-08-15. True: Anthropic's public submissions, policy proposals or statements to CAISI and/or the White House show it advocated a testing regime that applies more rigorous testing to frontier models than to off-frontier models, for example through capability or compute thresholds. Mostly true: the advocacy is documented with this tiering, but only before one of the two bodies, or a minor detail differs. Mixed: Anthropic advocated testing, but the tiered treatment of frontier versus off-frontier models is only partly or ambiguously present. Mostly false: Anthropic advocated testing, but its proposals did not distinguish frontier from off-frontier models. False: there is no such advocacy, or the documented proposals call for uniform or opposite treatment.

Set before the research began.

1. E1 Anthropic’s Advanced AI Framework Anthropic · June 2026 Own record

   “Models are covered when they have strong enough capabilities to pose catastrophic risks, with revenue- and spending-based thresholds to capture very large developers, based on our current understanding of training requirements for potentially dangerous models.”

2. E2 Anthropic’s Advanced AI Framework Anthropic · June 2026 Own record

   “Under these obligations, frontier developers would test their models for a defined set of catastrophic risks, show how those risks have been mitigated, publish what they find, and answer to a government agency with the authority to act on the results.”

3. E3 Anthropic’s Advanced AI Framework Anthropic · June 2026 Own record

   “The Agency should apply a fact-based review process consistently across developers, holding all covered models of equivalent capabilities to the same standards.”

4. E4 Anthropic's Advanced AI Framework Anthropic · June 2026 Own record

   “Models are covered when they have strong enough capabilities to pose catastrophic risks, with revenue- and spending-based thresholds to capture very large developers, based on our current understanding of training requirements for potentially dangerous models.”

5. E5 Anthropic's Advanced AI Framework Anthropic · June 2026 Own record

   “The following sections set out proposed obligations for the companies building the most capable AI models. Under these obligations, frontier developers would test their models for a defined set of catastrophic risks, show how those risks have been mitigated, publish what they find, and answer to a government agency with the authority to act on the results.”

6. E6 Anthropic's Advanced AI Framework Anthropic · June 2026 Own record

   “Capability threshold. Over time, the FLOP needed to train dangerous models may fall, and it may make sense to introduce a threshold based on capabilities rather than simply on training costs.”

7. E7 Policy on the AI Exponential Anthropic · 10 June 2026 Own record

   “These rules should apply only to models trained using more than 10²⁵ floating-point operations (FLOPs), developed by companies earning more than $500M in AI-related revenue or spending more than $1 billion on AI R&D.”

8. E8 Policy on the AI Exponential Anthropic · 10 June 2026 Own record

   “Frontier AI developers should have to test these models, be transparent to the public about their findings, submit them to independent evaluation, and maintain a robust security program.”

9. E9 AI policy Anthropic · 4 November 2026 Own record

   “The framework proposes a set of obligations for developers of the most capable models, who should test for these risks, engage independent evaluators, and disclose risk assessments and safety incidents on an ongoing basis.”

10. E10 Anthropic Response to OSTP RFI (March 2025) Anthropic · 6 March 2025 Own record

    “To establish effective government testing and evaluation for capabilities of powerful AI systems, the administration should:”

Anthropic's own policy documents propose testing obligations limited to the most capable models above compute and developer-size thresholds (E1, E5, E7, E8), but none of the evidence shows these proposals were made to CAISI or the White House. The one White House submission in the evidence (E10) recommends government testing of powerful AI systems without showing any frontier versus off-frontier tiering.