Amazon CEO Reportedly Flagged Anthropic Security Flaws

[Financial News]  


Analysts said Anthropic's complete suspension of its Claude Mythos 5 and Claude Fable 5 services on the 12th, under government orders, may have been triggered by the Amazon CEO's disclosure of the security vulnerabilities.
WSJ reported on the 13th, citing sources, that the Trump administration's "ban on all foreign users" came after Amazon CEO Andy Jassy discussed Anthropic's AI models with Treasury Secretary Scott Bessent and other U.S. officials.
According to the sources, Jassy told the administration that Amazon researchers had found information in Anthropic's Claude Fable 5 model that could help with cyberattacks, and suggested that restrictions might be necessary. Claude Fable 5 was designed to block commands that could be misused for cyber security or biological weapons, but Amazon researchers reportedly bypassed those safeguards by entering a series of prompts.
Soon after Jassy met with Bessent and other administration officials, the White House discussed possible responses, and security researchers verified Amazon's claims. The sources said they later concluded that the most direct way to address the risk was to prevent foreign governments, companies and individuals from accessing the AI models.
A senior White House official later said Trump approved the foreign-user ban for security reasons, despite concerns that it could hinder innovation.
According to WSJ, executives at technology companies regularly contact the administration to discuss the power of advanced AI tools.
The Trump administration had long distrusted Anthropic, believing the company had failed to properly manage security despite its leadership in the AI race. Sources said that concern was reaffirmed during a call on the 12th between Anthropic CEO Dario Amodei and administration officials.
Anthropic shut down all Claude Mythos and Claude Fable services on the 12th under government orders. The full suspension appears unavoidable. The company said many of its researchers are foreign nationals, and that the government's ban on foreign users effectively halted the company's work on its latest models.
Amazon.com, Inc., which provided the security vulnerability information, is also a major investor in Anthropic and supplies the semiconductors needed for data centers.
Anthropic countered that the vulnerabilities raised by Amazon were relatively basic and did not amount to a full system "Jailbreak." Some security experts agreed, WSJ reported.
WSJ said the unusually harsh scrutiny of Anthropic over security flaws that also affect other AI systems such as OpenAI's ChatGPT appears to be largely political retaliation and an attempt to rein in the company.
The Trump administration was uncomfortable with Anthropic's close ties to Democratic Party donors and its hiring of many people from the Biden administration. CEO Dario Amodei has also publicly criticized Trump.
In particular, when Anthropic opposed using its AI models in military operations, the Ministry of National Defense placed the company on a blacklist, and Anthropic is now pursuing legal action in response.
Kate E. Koren of the Center for Strategic and International Studies (CSIS) said the administration's security concerns were understandable, but added that the White House's strong hostility toward Anthropic likely had a major influence on the decision to impose the sweeping suspension.
Because Anthropic filed for an initial public offering (IPO) with the SEC on the 1st, the full service suspension could also hurt its listing prospects.
By contrast, OpenAI, which has recently fallen behind Anthropic and faced warning signs for its IPO prospects, is now likely to benefit. Unlike Anthropic, OpenAI has maintained close communication with the Trump administration and has taken a pro-Trump stance.
Amazon.com, Inc., which reported the security flaws, has also recently deepened its cooperation with OpenAI.

dympna@fnnews.com Song Kyung-jae Reporter