Anthropic removed the only binding safety commitment from its Responsible Scaling Policy. During the same period, its valuation soared from $4 billion to $965 billion, surpassing OpenAI.
Anthropic removed the binding commitment to pause development if models exceeded safety thresholds from its Responsible Scaling Policy version 3.0 in February 2026. The original commitment existed in version 1.0 from September 2023. It was replaced with self-assessed 'Frontier Safety Roadmaps.' In the same month, the company raised $30 billion, followed by an additional $65 billion in May, reaching a $965 billion valuation that surpassed OpenAI. On June 1, it filed a confidential draft registration for an IPO with the SEC.
Anthropic positioned itself as a 'responsible AI' company, emphasizing safety since its founding. However, as competitors accelerated, the company argued that unilaterally pausing development would not make the world safer. Chief Science Officer Jared Kaplan stated that unilateral commitments no longer made sense 'if competitors are blazing ahead.'
This event highlights the fundamental tension between safety and growth in the AI industry. Even the most safety-focused company abandoned its core safety mechanism under competitive pressure. The decision was made with seemingly sound reasoning and without coercion, making it particularly concerning. It suggests that voluntary commitments by individual companies are unlikely to survive without industry-wide coordination or regulation.
One comment notes that Anthropic set a higher bar for changing commitments in RSP v1.0 but actually lowered it in practice, and employees repeatedly described it as binding. If users relied on that promise to make decisions, they have a right to feel misled even if the policy change itself was justified. This argues that the legitimacy of the change and the breach of trust should be considered separately.