According to Katie Moussouris, CEO of Luta Security, who reviewed the report shared by Anthropic, Amazon researchers used a series of prompts to bypass Fable’s guardrails and access the underlying Mythos model. “They used prompts to generate output that could be used to find vulnerabilities and generate test scripts that could be turned into proof of concept code in a multistage process,” she told the Washington Examiner. “They were trying to bypass the Fable guardrails to get the underlying Mythos model to find vulnerabilities.”

Moussouris emphasized that the research was narrowly focused and did not uncover new risks in real-world code. “They tested this technique on known CVEs and on code they created with deliberate vulnerabilities inserted. No new vulnerabilities in real code were discussed in this paper,” she said.

She argued that the demonstrated capability is actually essential for cybersecurity defense. “Defenders need to use the latest AI models in exactly this way to keep up with attackers,” Moussouris said. “There’s no way to stop this from working without making the models less effective at finding bugs and generating test cases to test patches.”

Moussouris, who has previously advised the U.S. government on export controls, described the Trump administration’s response as a regulatory overreaction. “I see this as a misread of the research paper,” she said. “This was an example of regulatory overreaction that ends up being harmful to defense. Unintended consequences of regulation happen when there aren’t enough experts advising on how defense looks a lot like offense in cybersecurity.”

Amazon did not confirm that it performed jailbreak testing on Anthropic’s AI models, but did acknowledge that the company regularly consults the government on cybersecurity issues.

“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 spokesman told the Washington Examiner. “When they occur, we don’t share the details of these discussions.”

Commerce Secretary Howard Lutnick notified Anthropic CEO Dario Amodei on Friday that Fable 5 and Mythos 5 would be subject to export controls, according to an administration official. The restrictions prohibit access by customers outside the United States, along with foreign nationals within the country.

Anthropic said it believes the government’s decision was tied to a recently identified jailbreak technique, allowing users to bypass safety guardrails in the models. The company described the method as one that uncovered only a small number of previously known and relatively minor vulnerabilities in cybersecurity software.

“We reviewed a demonstration of this specific technique being used to identify a small number of previously known, minor vulnerabilities,” Anthropic said in a statement. “These vulnerabilities all appear relatively simple, and we have found that other publicly available models are able to discover them as well without requiring a bypass.”

The Washington Examiner reached out to Anthropic and the Commerce Department for comment.

Anthropic blocks most advanced AI models after US bars foreign use

Anthropic said it is complying with the government’s directive but strongly disagrees with the rationale behind the order.

“We are complying with the government’s legal directive and are removing access to Fable 5 and Mythos 5 for all users,” the company said. “However, we disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people.”