Amazon flagged concerns over Anthropic AI models before US crackdown
What's the story
Amazon CEO Andy Jassy was among tech leaders who warned senior Trump administration officials this week about potential security risks posed by Anthropic's most advanced AI models, according to Reuters. The warning comes as part of a broader effort by tech leaders to address potential cybersecurity vulnerabilities posed by these cutting-edge technologies.
Compliance action
'Fable' briefly went live before being taken down
Following the national security order from President Donald Trump's administration, Anthropic took the unusual step of shutting down its latest models worldwide. The company had previously warned about the hacking capabilities of its Mythos model and withheld it from wide release. However, earlier this week, they launched a public version called Fable with cybersecurity safeguards. This brief release was halted on Friday after receiving government concerns over potential bypassing of safeguards in place to prevent misuse for identifying cybersecurity vulnerabilities.
Security concerns
US believes 'jailbreak' exists to bypass safeguards
In a blog post, Anthropic said the US government believes there is a way to "jailbreak" its model and use it for finding cybersecurity holes. The bypass only found "minor" security flaws that other publicly available models can also detect, the company claimed. Responding to the Trump administration's order, Anthropic agreed to disable access to both Fable 5 and Mythos 5 globally for foreign nationals inside or outside the US.
Official statement
Amazon did not confirm discussions with government officials
An Amazon spokesperson said it is not uncommon for governments to seek their counsel on potential security risks, given their position as a leading cloud provider serving many private and public sector customers. However, the company did not confirm whether it had spoken to government officials about Anthropic's models. The spokesperson added that they do not share details of these discussions when they occur.
Regulatory action
Export control issued 'reluctantly': White House adviser
The US government restrictions were issued as an export control, according to Anthropic's blog post. White House adviser David Sacks said officials issued the export control "reluctantly" after Anthropic CEO Dario Amodei "refused" to "fix the jail break or de-deploy the model." Sacks, co-chair of Trump's Council of Advisors on Science and Technology and White House AI czar, expressed hope that Anthropic would remediate the safety issue so that Fable could be released again.
Expert opinion
Critics slam decision for impacting allies and adversaries alike
Experts supporting export controls on advanced AI models have questioned the Trump administration's decision as it affects both allied nations and adversaries. Jimmy Goodrich, a senior fellow at the University of California's Institute for Global Conflict and Cooperation, said: "This was not well thought-out." He added that it even prohibits Canadians and Brits working at Anthropic from conducting research and development.