Sam Altman's October 2026 comments that society should accept limited AI harm for broader access, paired with his statement that he agrees with Dario Amodei's call to decelerate the race toward the most powerful models, reveal a more complicated picture than the two companies' public rhetoric usually suggests. Anthropic has built much of its brand on pushing harder for AI oversight than its peers, while OpenAI has generally emphasized broad access and iterative deployment.
What's notable is where the two actually converge: both CEOs now appear to accept that the current pace of frontier model development carries risks neither company can fully control alone, even as they continue to disagree publicly on how much risk is acceptable in exchange for continued access and innovation. That's a meaningfully different conversation than the industry was having even a year earlier, when deceleration was a position associated almost exclusively with AI safety researchers rather than the labs themselves.