#

Key takeaways

• Anthropic CEO Dario Amodei called for slowing frontier AI development.

• The proposal includes embedded third-party evaluators and common safety standards.

• It also calls for broader global coordination.

• Anthropic said it would give outside evaluators access comparable to internal risk teams.

• OpenAI CEO Sam Altman said OpenAI would do the same.

What the proposal says

TechCrunch reported on September 12, 2026 that Anthropic CEO Dario Amodei called for slowing the pace of frontier AI development. The report says he proposed three broad measures. These are embedded third-party evaluators, coordination on common safety standards and limits, and global coordination.

The proposal is framed around development speed and oversight. It does not present a detailed policy package in the source text. Instead, it points to broad safeguards that could shape how advanced AI systems are reviewed and governed.

Third-party evaluators

One part of the proposal is to embed third-party evaluators. In practical terms, that means outside reviewers would have a role in assessing risk. The source does not define how those evaluators would be selected or what authority they would have.

Anthropic said it would unilaterally commit to giving outside evaluators access comparable to internal risk-assessment teams. That is a notable governance signal. It suggests a willingness to open internal review processes to external scrutiny.

Common standards and limits

The second measure is coordination on common safety standards and limits. This suggests a shared baseline for how frontier AI systems are assessed. The source does not list the standards themselves, so any specific framework would be an assumption.

This kind of coordination can matter because it reduces fragmentation. If multiple actors follow different rules, oversight becomes harder to compare. A common approach can make evaluation more consistent, at least in principle.

Global coordination

The third measure is global coordination. The source presents this as a broad goal rather than a detailed mechanism. It does not name institutions, agreements, or timelines.

Global coordination usually implies that frontier AI risks are not confined to one organization or one market. The source does not go further than that. So the safest reading is that Anthropic is asking for wider alignment, not a specific international program.

What OpenAI said

The report says OpenAI CEO Sam Altman said OpenAI would also give outside evaluators access comparable to internal risk-assessment teams. That places another major AI company in the same general direction on evaluator access.

The source does not say whether OpenAI endorsed the full proposal. It only states that Altman made the access commitment. Readers should treat that as a separate statement unless more detail is provided.

Why this matters for AI governance

The main theme here is governance, not product launch. The proposal focuses on how frontier AI should be reviewed before and during deployment. That makes it relevant to anyone tracking operational risk in advanced AI systems.

The source does not provide evidence of a new law, a binding standard, or an enforcement mechanism. It describes a proposal and a voluntary commitment. That distinction matters because voluntary commitments can guide behavior without creating formal obligations.

Operational considerations

For organizations watching frontier AI development, the report highlights a few practical questions. How much access should outside evaluators receive? How should safety standards be compared across teams? And what level of coordination is realistic across different actors?

The source does not answer those questions. Still, it shows that access, standards, and coordination are central to the current governance debate. Readers can treat those as the main operational themes in the report.

Morocco relevance

The source reports no Morocco-specific policy, partnership, or deployment detail. For readers in Morocco, the global lesson is that AI governance discussions may increasingly focus on external review and shared safety rules.

Bottom line

Anthropic's CEO is pushing for a slower and more coordinated approach to frontier AI development. The proposal centers on third-party evaluators, common safety standards, and global coordination.

Anthropic also said it would give outside evaluators access comparable to internal risk teams. OpenAI's CEO said OpenAI would do the same. The report presents these as governance signals, not as final policy outcomes.