TechFlow News, June 14: David Sacks, the White House’s AI and Cryptocurrency Affairs Lead, stated that Anthropic’s Mythos series commercial model “Fable,” released this week, includes safety guardrails—but users can still bypass these restrictions via specific jailbreaking methods to access Mythos’s advanced cyberattack capabilities.
Sacks noted that a partner trusted by both Anthropic and the U.S. government discovered the vulnerability during testing. The U.S. government subsequently requested Anthropic CEO Dario Amodei either fix the flaw or temporarily take the model offline—a request Anthropic refused.
Anthropic maintains that the vulnerability is not severe enough to impair normal product functionality. However, the U.S. government and its relevant partners disagree.
Sacks stated that while Anthropic has long emphasized AI safety as its top priority, it prioritized maintaining consumer-grade model services in this instance. In response, the U.S. government has imposed export controls on Anthropic and indicated it will consider lifting the restrictions once the issue is resolved.
