Smart voices in tech react to return of Anthropic's Fable and Mythos

Anthropic Restores Access to Fable 5 and Mythos 5 Following Commerce Department Decision

Anthropic announced Tuesday that the Department of Commerce has lifted export controls on its Fable 5 model, allowing the company to restore general access. The AI company also confirmed that access to its more advanced Mythos 5 model has been reinstated for a vetted group of organizations after receiving separate US government approval the prior week.

The restrictions originated in June when the White House directed limits on the models. Officials cited concerns that a specific jailbreak technique identified in Fable 5 could enable adversaries to use the systems for cyberattacks. Anthropic stated at the time that the action stemmed from a misunderstanding regarding the jailbreak and dispatched senior executives to Washington for discussions. The company indicated that restored access includes additional safeguards developed during the review period.

The episode has drawn attention to procedures for reviewing advanced AI systems with potential applications in cybersecurity and other sensitive domains. Several technology figures commented on the outcome and its implications for future model releases.

Alex Stamos, a Stanford University lecturer and former chief security officer at Facebook, posted on X that the suspension represented a setback for US interests. He argued that limits on American AI models could encourage cybersecurity firms to adopt Chinese alternatives, stating that security companies and startups providing services to others would now be driven toward those options. Stamos described the development as a gain for PRC laboratories and noted that Anthropic's statements emphasized its ongoing focus on safety, initial handling of the issue, agreement from US government technical experts, and plans to develop clearer standards for communication.

Dean W. Ball, a senior fellow at the Foundation for American Innovation and former Trump administration AI official who is joining OpenAI, posted on X welcoming the return of access to both models. He observed that questions remain about the specific safety commitments Anthropic made during negotiations and whether comparable standards will apply to other frontier systems. Ball wrote that continued opacity around such agreements would not support a stable and investable industry over time, while adding that the government does not need to resolve every detail immediately and that a two-week review period is not unreasonable. He described the lifting of the restrictions as progress.

Aaron Levie, chief executive of Box, posted on X that the resolution could establish a precedent for reviews of future frontier models, especially those possessing advanced capabilities in coding, cyber operations, and biosecurity. He noted that assessments of risk and exploitability involve substantial subjectivity. Levie stated that companies will likely operate under a framework requiring ongoing judgment and exchanges between AI laboratories and government agencies for major releases. He expressed support for Anthropic's proposal of a shared industry framework and increased collaboration with the US government but cautioned that such reviews will continue to depend on significant discretionary evaluation. Levie added that the goal should be an efficient process.

The comments reflect differing emphases on competitiveness, procedural clarity, and the balance between security reviews and development timelines. Anthropic has indicated it will work with other companies on standardized approaches to safety evaluations in coordination with federal agencies.