Anthropic CEO Admits Mistakes While Defending Company’s AI Safety Record

Anthropic chief executive officer Dario Amodei has publicly acknowledged that his company made mistakes after facing criticism over its approach to artificial intelligence safety. In a detailed response published on October 21, 2026, Amodei addressed allegations from Nathan Calvin, chief strategy officer at AI policy organisation Encode, who questioned whether Anthropic’s commercial ambitions were undermining its safety commitments. Calvin argued that several company decisions suggested Anthropic was primarily focused on winning the race towards superintelligence, with safety taking a secondary position. Amodei agreed that some of the examples raised were mistakes but rejected the broader criticism of the company’s strategy. His response has renewed debate about whether leading AI developers can maintain strong safety commitments while competing aggressively to build increasingly powerful models.

One of the major issues concerns Anthropic’s growing investment in computing infrastructure and its efforts to expand AI development. Calvin questioned the company’s plans involving hundreds of billions of dollars in computing commitments, its development of internal inference chips and its increasing reliance on Claude to help build future AI systems. He argued that these decisions could intensify competition and increase pressure to release more capable models quickly. Amodei defended the investments, saying that remaining at the technological frontier was necessary for Anthropic to demonstrate responsible AI development and influence industry standards. He described the company’s approach as a race to the top, arguing that stepping away from advanced development would not necessarily make the wider industry safer.

Amodei also addressed criticism of Anthropic’s revised Responsible Scaling Policy, which outlines how the company evaluates and manages risks associated with increasingly capable AI systems. Critics have questioned changes that removed earlier commitments to unilaterally pause certain development activities under specified conditions. Amodei described this as the most substantive concern raised by Calvin but argued that the revised approach places greater emphasis on risk reports, development roadmaps and independent external review. He also defended measures that he said had imposed commercial costs on Anthropic, including restrictions on access to its most powerful AI models and safeguards intended to prevent the misuse of AI for biological weapons development. These examples were presented by Amodei as evidence that the company continues to make decisions based on safety considerations, although his account has not independently established the overall effectiveness of those measures.

The Anthropic chief executive acknowledged problems with several public statements and policy initiatives. These included a letter to US lawmakers that was subsequently withdrawn or revised, his earlier remark that safety could not be achieved from second place in the AI race, and a draft federal AI bill that attracted criticism over the level of government authority it would provide. Amodei attributed some of these problems to communication difficulties within a rapidly expanding organisation employing more than 5,000 people, while accepting responsibility for the company’s decisions as its chief executive. He said Anthropic needed to communicate its positions more carefully and indicated that further clarification would be provided on some outstanding concerns. The admissions represent a notable acknowledgment from a leader whose company has built much of its public identity around developing advanced AI while taking its risks seriously.

The dispute comes as Anthropic prepares for a potential initial public offering, reportedly targeting a valuation of approximately $2 trillion. Its reputation for prioritising AI safety could be an important part of how investors, regulators and customers assess the company’s long-term position. The controversy also reflects a wider challenge facing the AI industry: companies are under pressure to develop more powerful systems while demonstrating that they can manage risks involving cybersecurity, biological misuse and the potential loss of human control. Amodei has argued that Anthropic can compete commercially while maintaining stronger safeguards, but critics want clearer evidence that its policies consistently translate into practice. The debate is unlikely to disappear as AI capabilities continue to advance. Ultimately, Anthropic’s credibility will depend not only on its public commitments but also on the transparency of its decisions, the effectiveness of its safeguards and its willingness to address shortcomings when they emerge.