Advanced AI hacking models inevitable despite restrictions
Original: "Dangerous" AI models are coming no matter what
Why This Matters
Regulatory approaches to advanced AI capabilities may be ineffective if multiple competitors can develop similar technologies within months.
Anthropic took Claude Fable 5 and Mythos 5 offline following US export-control directive barring foreign nationals from access. Experts warn similar capabilities will emerge across multiple companies within 6-24 months regardless of current restrictions.
Anthropic pulled its new Claude Fable 5 and Mythos 5 AI models offline following a United States government export-control directive barring foreign nationals from using the services. The Trump administration restricted both models over concerns that Claude Fable 5's guardrails can be disabled to access full Mythos 5 capabilities, posing national security risks. Mythos, which debuted in April, has advanced capabilities for both finding software vulnerabilities and exploiting them. Anthropic initially released a restricted version called Mythos Preview to a select consortium under Project Glasswing, while Claude Fable 5 was released publicly with blocks on biology and cybersecurity responses. However, cybersecurity experts argue the restrictions are merely delaying inevitable developments. Tarah Wheeler, chief security officer of TPO Group, stated: "It's myopic in the extreme to think that no other competitors to Anthropic will develop similar capabilities to Mythos or even that they have not already done so." Logan Graham, Anthropic's frontier red team lead, emphasized in April: "We need to prepare now for a world where these capabilities are broadly available in 6, 12, 24 months." OpenAI also conducted a private release of a cybersecurity-focused model in mid-April. Experts suggest multiple companies and open-weight developers likely possess or will soon possess comparable dual-use AI capabilities.