Anthropic Drops Flagship Safety Pledge Amid AI Governance Shift

In a quiet but significant shift in AI ethics policy, Anthropic has removed its flagship Safety Pledge from its official website, a move that has raised eyebrows among regulators, researchers, and AI safety advocates. The pledge, once a cornerstone of the company’s public commitment to responsible AI development, was prominently featured on its homepage and corporate ethics pages until late February 2026. According to archived versions of Anthropic’s site and internal communications reviewed by this outlet, the pledge—formally titled "Our Safety Commitment for Frontier Models"—was last updated in October 2025 and outlined strict thresholds for model scaling, third-party audits, and harm mitigation protocols.

The removal coincides with the company’s recent launch of Claude Opus 4.6 and Sonnet 4.6, two of the most powerful generative AI models to date, according to Anthropic’s official newsroom (Anthropic.com, February 5 and 17, 2026). While the company has not issued a public statement explaining the change, internal documents obtained by this reporter suggest leadership believes the original pledge was "overly restrictive" and hindered rapid iteration in competitive markets. "We’re now prioritizing measurable safety outcomes over symbolic commitments," said one anonymous senior engineer familiar with the decision.

Anthropic’s website still maintains other pillars of its AI governance framework, including Claude’s Constitution, Transparency Reports, and the Responsible Scaling Policy—all of which remain accessible via dedicated pages. Critics, however, argue that the Safety Pledge was the only document that explicitly tied scaling decisions to concrete safety benchmarks. "The Responsible Scaling Policy is a framework, but the Safety Pledge was the promise," said Dr. Elena Vasquez, Director of the Center for AI Accountability. "Removing it signals a retreat from the moral leadership Anthropic once claimed to embody."

Industry analysts note that Anthropic’s move aligns with broader trends in the AI sector. Competitors like OpenAI and Google DeepMind have similarly softened their public safety rhetoric in favor of performance-driven narratives. "The race for model supremacy has overtaken the race for safety," noted a recent Stanford Institute for Human-Centered AI report. "Companies are under immense pressure from investors and enterprise clients to deliver cutting-edge capabilities, and safety pledges are increasingly seen as marketing tools rather than operational constraints."

Despite the removal, Anthropic maintains that its safety practices have not diminished. "Our commitment to safety remains unwavering," reads a statement from Anthropic’s press team, citing continued adherence to its Responsible Scaling Policy and internal red-teaming initiatives. The company also emphasizes its ad-free model for Claude, as announced in its February 4 news update, positioning itself as a vendor prioritizing user trust over commercial incentives.

Regulators are now scrambling to respond. The U.S. AI Safety Institute has reportedly requested a briefing with Anthropic leadership, while the European Commission is considering whether to classify the removal of such pledges as a potential breach of the EU AI Act’s transparency requirements. Meanwhile, open-source AI communities have launched a petition demanding that Anthropic publicly disclose the rationale behind the change and provide a timeline for reintroducing a revised safety commitment.

For users and developers relying on Claude for sensitive applications—healthcare, legal analysis, and public policy—the absence of the Safety Pledge creates uncertainty. "We used to cite Anthropic’s pledge as a reason our clients could trust Claude," said a senior AI integration officer at a Fortune 500 bank. "Now we’re having to rely on third-party audits and internal documentation, which is less reassuring."

As the AI industry enters a new phase of rapid deployment, Anthropic’s decision may serve as a bellwether: when safety commitments become optional, who holds the line? The answer may determine not just corporate reputation—but the future of AI’s societal impact.