Amazon probably raised worries about Anthropic AI models
Amazon CEO Andy Jassy was among tech leaders who expressed concerns to senior officials in the Trump administration. This week, a source with knowledge of the situation informed about the security risks associated with Anthropic’s most advanced AI models. Jassy’s involvement highlights the significant action taken by Anthropic on Friday to globally discontinue its latest models in response to national security directives from President Donald Trump’s administration. The San Francisco-based AI startup has confidentially filed for a US initial public offering. It had previously cautioned about the hacking capabilities of its Mythos model and had withheld it from a broad release. However, earlier this week, Anthropic introduced a public version named Fable, which it characterised as having cybersecurity safeguards.
That brief release concluded on Friday. In a recent blog post, Anthropic reported that the US government has indicated it believes there is a way to bypass, or “jailbreak,” a safeguard designed to prevent the model from being used to identify cybersecurity vulnerabilities. The bypass identified only “minor” security flaws that other publicly available models can also detect, according to Anthropic’s blog post. The Trump administration mandated that Anthropic restrict access to its latest models, Fable 5 and Mythos 5, for all foreign nationals, regardless of their location, as stated by the company. In response, Anthropic announced its decision to disable access to the models on a global scale. Amazon has not verified if it engaged in discussions with government officials regarding Anthropic’s models. “As a leading cloud provider that serves a large number of private and public sector customers, it’s not uncommon for governments to seek our counsel on potential security risks,” an Amazon spokesperson said. When they occur, we refrain from disclosing the specifics of these discussions.
The Information, a technology news outlet, earlier on Saturday reported on Jassy’s concerns. The Information, referencing a US official, subsequently reported that the administration was not expected to compel other AI firms to adhere to restrictions akin to those imposed on Anthropic. The US government restrictions came in the form of an export control, Anthropic said in its blog post. The US Commerce Department’s Bureau of Industry and Security, responsible for overseeing export controls, has not yet provided a response to the request for comment. Officials implemented the export control “reluctantly” following Anthropic CEO Dario Amodei’s “refusal” to “fix the jail break or de-deploy the model,” as noted by White House adviser David Sacks in a social media post on Saturday.
“The Admin’s hope now is that Anthropic remediates the safety issue, the export control is lifted, and Fable goes back into general release,” wrote Sacks. Some experts who support export controls on advanced AI models found the actions of the Trump administration to be perplexing, as they impact both allied nations and adversaries. “This was not well thought-out,” said Jimmy Goodrich. It even prohibits Canadians and Brits working at Anthropic from engaging in research and development. The order arrived at a time when a prior disagreement between officials from the Trump administration and Anthropic appeared to be calming in various sectors of the US government.
