Anthropic is dropping its signature safety pledge amid a heated AI race

In a significant shift, Anthropic, the AI startup founded by ex-OpenAI employees, has announced a change to its foundational safety commitments. This move comes amid increasing competition within the AI sector and a notable absence of government regulation. On Tuesday, Anthropic revealed that it will no longer adhere to its previous promise to pause or delay the deployment of new models when advancements would exceed its safety measures. This decision allows the company to operate with fewer restrictions on safety concerns at a time when its key product, the chatbot Claude, is making waves in financial markets and raising alarms about the future of software. The company has introduced a new set of safety guidelines known as the Responsible Scaling Policy, which distinguishes between its own practices and those of the broader AI industry. This policy, inspired by U.S. biosafety standards, reflects Anthropic's view of the current competitive landscape. Jared Kaplan, the company’s chief science officer, shared with Time Magazine that the rapid pace of AI advancements made it impractical for them to halt their model training efforts. "We felt that it wouldn't actually help anyone for us to stop training AI models," Kaplan explained. While the new guidelines still include a commitment to delay the release of highly capable AI models in specific situations, the criteria for doing so have been narrowed. Anthropic highlighted an overall unfavorable political climate towards regulation as a factor influencing this decision. Despite some success in advocating for AI regulations at the state level, the company has yet to see substantial action at the federal level. Anthropic remains dedicated to promoting effective government engagement on AI safety, emphasizing the importance of evidence-based discussions that take into account national security, economic competitiveness, and public trust. However, the company acknowledges that meaningful change is a long-term endeavor. The scaling policy was intended as a dynamic guideline, evolving with the industry. Dario Amodei, the CEO, previously articulated that safety measures were crucial to mitigate the risks associated with powerful AI technologies. He referenced the well-known phrase, "With great power comes great responsibility," to underline the inherent risks of powerful AI models. Anthropic also noted that risks categorized as ASL-4 and above cannot be managed by a single entity alone. Amodei has consistently pointed to the company’s commitment to safety, recalling their decision to delay the launch of Claude in the summer of 2022 due to concerns about insufficient safeguards for such a groundbreaking technology. Reflecting on this decision, Amodei admitted that it likely cost them a competitive edge in the consumer AI space, especially after OpenAI launched ChatGPT in November 2022. He asserted that their cautious approach was a testament to their dedication to safety, despite criticisms that their advocacy for regulation serves to stifle competition. In a recent discussion, Amodei addressed claims that the company benefits from being the only one advocating for certain regulations, indicating that such a perspective oversimplifies the complexities involved in the industry. He encouraged stakeholders to judge Anthropic by its actions rather than assumptions about its motives.