Researchers say the Fable 5 controversy started not with a jailbreak but with 'fix this code'

The U.S. government issued export control guidance restricting access to Anthropic's Fable 5 and Mythos 5 models citing national security concerns, but security experts have pushed back against this.

Katie Moussouris, CEO of Luta Security, argued that the models were not subjected to a 'Jailbreak' attack. Instead, the issue began when researchers input code containing vulnerabilities and then asked the model to "fix this code," and the model responded. She explained this was a normal 'find, fix, and test loop' process used for security review and patch verification.

The key points of contention are as follows:

• Export control background: The U.S. government restricted access over security concerns that the models could be exploited for attacks leveraging vulnerabilities, and Anthropic blocked access for all customers in order to comply with the regulations.
• Security experts' counterargument: More than 100 security leaders warn that these restrictions harm defenders more than attackers. Their position is that if AI-assisted bug discovery and patch verification capabilities are restricted, the overall cybersecurity ecosystem could be weakened.
• Technical point of contention: Researchers emphasize that the model did not bypass security guardrails, but rather generated the security vulnerability and test scripts while simply responding to a code-fix request.