Changes at Anthropic Following Government Scrutiny
In a surprising turn of events, Dario Amodei, the CEO of Anthropic, has been replaced by co-founder Tom Brown during a critical meeting at the White House. It appears that Amodei’s behavior had been described as, well, a bit eccentric.
After the U.S. government enforced strict export controls on Anthropic’s latest AI models, “Mythos” and “Fable,” Amodei and other senior staff members quickly traveled to Washington. But it seems Amodei was hard to engage with, brushing aside officials’ concerns, as reported by a source.
In contrast, Brown is perceived as more relatable, according to someone familiar with the situation.
Recent conversations between Anthropic and the Trump administration seem to be improving since this leadership change. Brown and Sarah Heck, the company’s head of public policy, have been active in discussions that are reportedly going well.
Neither the White House nor Anthropic has responded to requests for comments.
Earlier this month, the government implemented export controls on Anthropic’s powerful new AI model after warnings were issued by Amazon CEO Andy Jassy regarding potential safety issues.
In response to these restrictions, Anthropic took its models offline on June 12, asserting that it was necessary to comply with the new guidelines.
According to reports, the National Security Agency tested Anthropic’s AI tools and found that “Mythos” could identify vulnerabilities in secure government systems within a matter of hours. However, this doesn’t necessarily imply that the bots can exploit those vulnerabilities that quickly.
These tests are part of an initiative called Project Glasswing, where Anthropic collaborates with other organizations to assess AI models for safety concerns following alarming reports about the potential dangers posed by “Mythos” and “Fable.”
While talks between Anthropic and Trump officials appear to be progressing, the timeline for lifting the export restrictions remains uncertain.
Some discussions reportedly centered on what evidence Anthropic could provide to mitigate government worries about the security risks of new AI models.
Commerce Secretary Howard Lutnick, who is involved in these talks due to his department’s role in overseeing export controls, is under pressure from a bipartisan group of lawmakers. They set a June 26 deadline for him to respond to a series of questions regarding the way forward.
Whether the department will answer the letter by the deadline is still unclear.
Reportedly, Anthropic has been keen to improve its relationship with the government and address security issues, and has made recent commitments to collaborate more closely with the White House regarding its proposal to Lutnick.
The strain between Anthropic and the U.S. government peaked earlier this year when Amodei declined to grant the Pentagon unrestricted access to the AI bots during contract negotiations, advocating for the limitation of their use in mass surveillance and weaponry.
After the government imposed export restrictions, Anthropic argued they were merely highlighting potential safety risks associated with the bots, denouncing the measures as unwarranted.
Nonetheless, it seems White House officials were displeased that Anthropic downplayed these risks, despite Amodei’s warnings about other AI models’ potentially catastrophic effects over the years.
