Amazon research triggers White House ban on Anthropic's Fable 5
Amazon's cybersecurity research led to the White House blocking Anthropic's Fable 5 export...
Amazon’s cybersecurity research uncovered vulnerabilities in Anthropic’s Fable 5 model, where specific prompts could extract information potentially usable in cyberattacks. This prompted CEO Andy Jassy to brief White House officials, leading to an export control directive that restricted access for foreign nationals. The directive forced Anthropic to cut off access to Fable 5 and Mythos 5, complicating operations for the company’s foreign-born researchers.
Anthropic pushed back on the government’s ‘jailbreak’ framing, arguing that similar vulnerabilities exist in publicly available models like GPT 5.5. Security researchers, such as Katie Moussouris of LutaSecurity, supported this view, suggesting the paper did not demonstrate a true jailbreak. Meanwhile, tensions between Anthropic and the Trump administration have escalated, with federal agencies previously instructed to stop using Anthropic’s AI due to concerns over surveillance and autonomous weapons.
- Amazon’s research allegedly exposed vulnerabilities in Anthropic’s Fable 5, enabling the White House to block its export via an export control directive.
- Anthropic disputes the ‘jailbreak’ claim, stating similar issues exist in models like GPT 5.5, and researchers argue the paper did not demonstrate a true exploit.
- Tensions between Anthropic and the Trump administration have flared, with federal agencies previously banned from using Anthropic’s AI over ethical concerns.
Why It Matters
This clash highlights escalating government oversight of AI models, balancing security risks with innovation and access to critical research tools.