Anthropic CEO Dario Amodei has defended the artificial intelligence company’s approach to safety following pointed criticism from civil society groups and policy advocates over a recently proposed federal AI safety bill. The debate follows reports that Anthropic shared a draft federal safety bill with United States lawmakers, drawing swift backlash from tech policy watchdogs regarding specific provisions, including a high threshold for government intervention requiring 50 deaths or one billion dollars in damages before regulators can step in.
Nathan Calvin, chief strategy officer at the AI advocacy group Encode AI, voiced concerns on social media questioning whether Anthropic was prioritizing its ambition to build powerful AI models over genuine safety commitments. Calvin pointed to several recent company actions, such as investments in computing infrastructure, custom chip development, changes to its Responsible Scaling Policy, and the federal bill itself, arguing these steps aligned more with winning the race to develop advanced AI than ensuring safety.
In response to the critique, Amodei stated that while any corporate decision can be framed negatively, Anthropic has repeatedly made commercially costly choices to mitigate risks associated with advanced technology. He admitted that the company had made mistakes along the way and conceded that some of Calvin's criticisms carried valid points.
Amodei explained that Anthropic’s overarching strategy involves pursuing a race to the top, where developing advanced AI responsibly encourages stronger safety practices across the entire industry. He argued that staying at the technological frontier grants the company vital influence to steer the sector toward safe development, though he acknowledged that holding back frontier capabilities also carries broader costs by delaying potential benefits in fields like biology.
To support his position, Amodei highlighted multiple decisions that involved financial or business sacrifices, including supporting California’s SB 53 AI safety law, maintaining strict safeguards against biological threats, and delaying model rollouts until enhanced cybersecurity measures were established. He also noted the company’s support for export controls on advanced chips to China and restrictions on selling services to Chinese government-controlled entities.
The public discussion underscores a broader industry-wide tension over whether private companies can successfully scale increasingly powerful artificial intelligence systems while keeping safety oversight robust. While Amodei contends that maintaining a leadership position allows Anthropic to advocate effectively for national security and responsible governance, critics maintain that external scrutiny remains essential to ensure commercial incentives do not compromise public safety.