Aidan Clark, OpenAI's vice president of research (training), posted on X that he keeps hearing from Anthropic colleagues the sentiment that only Anthropic is trustworthy enough to build AI. He's not happy about it, arguing that having multiple companies working toward AGI is a good thing — going it alone almost guarantees picking the wrong path.
https://twitter.com/_aidan_clark_/status/2052089187659346047
Clark clarified in replies that these are hearsay, and he'd be happy if someone proved it was a misunderstanding. He also noted that his colleagues at OpenAI don't shy away from publicly acknowledging Anthropic's work.
The post drew responses from several Anthropic employees. One user claiming to work at Anthropic replied that they chose the company because they trust it more, but that's different from saying "only we can be trusted." Another user, Rohan Seth, directly denied the claim: "That's not our thinking," and deleted an earlier aggressive reply. Steven Adler offered a distinction: what Clark heard might not be "Anthropic thinks no one else should build AI" but rather "Anthropic doesn't trust OpenAI specifically" — two different things. Clark acknowledged the distinction, allowing that he might have heard the latter framed as the former, but added: "Even if it's just OpenAI, that's not much better."
Clark later softened his stance, saying maybe few people are extreme enough to think "only Anthropic should build AI," but he believes a milder version exists. He praised the response from the Anthropic side, both in public replies and private messages, calling them friendly.