Full Breakdown
Anthropic Overhauls Safety Policy Amid AI Race
2/25/2026, 5:58:17 AM
Shift in Safety Commitment
Anthropic, a leading AI company, has announced a significant change to its flagship safety policy, the Responsible Scaling Policy (RSP). Initially, the RSP included a commitment to refrain from training AI systems unless adequate safety measures were guaranteed. However, company officials have confirmed that this central pledge has been dropped. Jared Kaplan, Anthropic's chief science officer, explained that the decision was made in light of the rapid advancements in AI technology and the competitive landscape, stating, “We didn't really feel, with the rapid advance of AI, that it made sense for us to make unilateral commitments … if competitors are blazing ahead.”
New Policy Framework
The revised RSP emphasizes transparency regarding safety risks associated with AI and commits to matching or exceeding the safety efforts of competitors. The new policy allows for the continuation of AI development even when safety measures are not fully in place, provided that the company believes it can manage the associated risks. This shift reflects a pragmatic response to the evolving political and scientific environment, as Kaplan noted that the anticipated regulatory frameworks have not materialized, and competition in AI has intensified.
Industry Implications
The change has raised concerns among experts regarding the potential risks of AI development. Chris Painter, director of policy at the nonprofit METR, remarked that the decision indicates a shift towards "triage mode" in safety planning, suggesting that current methods to assess and mitigate risks are lagging behind technological advancements. Painter expressed concern that the removal of strict thresholds for halting AI development could lead to a gradual increase in danger without clear warning signs.
Official Statements & Responses
Anthropic maintains that the new RSP retains key elements of the original policy, including commitments to publish detailed “Frontier Safety Roadmaps” and “Risk Reports” every three to six months. These documents aim to outline safety goals and assess overall risk levels associated with AI capabilities. Kaplan emphasized that the company remains committed to developing AI safely, asserting, “If all of our competitors are transparently doing the right thing when it comes to catastrophic risk, we are committed to doing as well or better.”
Criticism & Opposition
Despite the company's reassurances, critics like Painter warn that the revised policy could undermine efforts to manage AI risks effectively. The absence of a clear stopping point for AI development raises fears of a "frog-boiling" effect, where dangers escalate gradually without immediate recognition. This perspective highlights the ongoing tension between innovation and safety in the rapidly evolving AI landscape.
Conclusion
Anthropic's decision to overhaul its safety policy marks a pivotal moment in the AI industry, reflecting both the pressures of competition and the complexities of ensuring safety in advanced technologies. As the company navigates this new path, the implications for AI governance and risk management remain a critical area of concern for stakeholders across the sector.
