Anthropic Revises Key AI Safety Pledge Amid Industry Pressure

Anthropic, one of the leading AI research firms behind the Claude family of models, has formally retired its flagship safety pledge from 2023 — a bold promise that the company would not train future AI systems unless it could guarantee in advance that its safety protocols were sufficient. According to an exclusive report by TIME, CEO Dario Amodei approved a revised internal policy that replaces the absolute prohibition with a more nuanced risk-assessment framework, allowing continued development under enhanced monitoring rather than a hard stop.

The original pledge, announced in early 2023, was hailed as a pioneering effort in responsible AI development, setting Anthropic apart from competitors like OpenAI and Meta, who had not made similarly binding commitments. At the time, the pledge was viewed as a moral and operational bulwark against unchecked scaling of large language models, particularly amid growing concerns over alignment, bias, and potential misuse. The decision to rescind it, however, has sparked intense debate within the AI ethics community and among investors.

Internal documents reviewed by TIME indicate that the change was driven by a combination of competitive pressures, evolving regulatory landscapes, and internal assessments that the original policy was too rigid to be practically enforceable. While Anthropic continues to emphasize its commitment to safety, the new framework prioritizes iterative risk evaluation, third-party audits, and real-time monitoring over pre-training guarantees. This shift mirrors broader industry trends, where companies are increasingly adopting adaptive governance models rather than static thresholds.

On Hacker News, the announcement drew over 150 comments, with many technical contributors expressing concern that the move undermines public trust. "This wasn’t just a policy — it was a signal to the world that Anthropic took safety seriously," wrote one user. Others noted that the original pledge may have been more symbolic than operational, as the definition of "adequate safety measures" was never publicly standardized. Critics argue that without transparent, measurable criteria, the revised policy lacks accountability.

Meanwhile, industry analysts suggest the change reflects a pragmatic response to global AI competition. With China, the U.S., and the EU all racing to establish regulatory frameworks, Anthropic may be positioning itself to comply with future legislation without sacrificing its development cadence. The company has not disclosed specific metrics for its new safety evaluation process, but insiders say it now includes continuous model behavior tracking, red-teaming exercises, and mandatory review by an expanded internal Safety Oversight Committee.

Anthropic’s decision comes as other AI firms, including Google DeepMind and xAI, have also moved away from public safety pledges in favor of internal governance. The trend raises questions about whether voluntary commitments can sustain public confidence in the absence of enforceable regulation. Advocacy groups such as the Center for AI Safety and the Algorithmic Justice League have called for federal legislation to codify safety standards, warning that corporate self-regulation is insufficient.

Despite the controversy, Anthropic maintains that its revised approach is more robust. "We’re not abandoning safety — we’re evolving it," said a company spokesperson in a statement to Reuters. "The world of AI is moving too fast for binary yes/no decisions. Our new system allows us to respond dynamically to emerging risks while continuing to deliver value to our users."

As the AI industry enters a new phase of rapid deployment and regulatory scrutiny, Anthropic’s policy shift may serve as a bellwether for the broader sector. Whether this evolution enhances safety or erodes accountability remains to be seen — but one thing is clear: the era of absolute pledges in AI development may be over.