Anthropic chief executive Dario Amodei has pledged that his company will slow down the development and release of new artificial intelligence systems as much as necessary to ensure their safety. He made the commitment while addressing the United Nations Security Council, marking one of the most prominent stages on which an AI lab leader has discussed safety trade-offs.

Amodei said Anthropic would not race ahead with releases if doing so risked public safety, framing caution as a deliberate company policy rather than a reluctant concession. The remarks come as frontier AI models grow more capable and governments step up efforts to understand their risks.

Safety As A Deliberate Policy

The Anthropic CEO has long argued that AI labs should accept slower progress in exchange for stronger safety guarantees. Speaking before the Security Council, he tied that position to global stability, suggesting that unchecked AI development could carry consequences far beyond the technology industry.

Anthropic, which builds the Claude family of models, has positioned itself as the safety-first alternative to competitors racing toward more powerful systems. Amodei’s statement reinforces that branding at a moment when the company is also pushing its models into new scientific and enterprise applications.

The Global AI Governance Debate

The address landed in the middle of a wider dispute over who should set AI safety rules. Calls from OpenAI’s Sam Altman, Amodei himself and Hugging Face’s Clement Delangue for international risk-evaluation standards have so far been rebuffed by Washington, which has resisted binding global frameworks for AI oversight.

By taking the pledge to the Security Council, Amodei effectively made the case that AI safety is not merely a commercial or technical question but a matter of international concern. The Council, traditionally focused on conflicts and weapons, is increasingly being drawn into debates about emerging technologies with cross-border impact.

What Slowing Down Means In Practice

For Anthropic, the commitment implies longer internal testing cycles, staged rollouts and a willingness to delay launches when safety evaluations raise red flags. Critics of the AI industry have questioned whether such pledges can survive competitive pressure, but Amodei’s public, on-the-record statement at the UN sets a benchmark his company will now be measured against.

With regulators around the world still drafting AI rules, the question of whether self-restraint by leading labs is enough — or whether enforceable standards are needed — remains open. Amodei’s appearance suggests the industry’s own leaders want that conversation to happen at the highest level.