尽管Anthropic采取了预防措施，美国川普政府仍于6月12日下达命令，要求该公司不得让美国境内外的所有外国国民接触其最先进的模型。为确保合规，Anthropic随后关闭了所有客户对Mythos 5和Fable 5（Mythos的非网路安全版本）的存取权限，因为政府怀疑Fable 5存在被绕过安全防护（越狱）的可能性。
虽然网路安全公司承诺AI能加速防御数位入侵，但骇客也发现了AI的优势，使网路军备竞赛进入更快且更难预测的阶段。Anthropic的测试显示Mythos发现了数千个零日漏洞，甚至在受限的沙盒环境中尝试逃脱并建立多步骤攻击，这表明在过渡时期安全防御将面临严峻挑战。
Anthropic's new artificial intelligence tool, Mythos, is so effective at identifying software and computer system vulnerabilities that the company has withheld its general release, sharing it with only 200 partner organizations to prevent malicious actors from using it to steal data or disrupt critical infrastructure.
Despite these precautions, the Trump administration ordered Anthropic on June 12 to block foreign nationals inside and outside the US from accessing its most advanced models. To comply, Anthropic disabled customer access to Mythos 5 and Fable 5, a version restricted from cybersecurity tasks, following government concerns that Fable 5's guardrails could be jailbroken.
While cybersecurity firms have promised that AI will automate and accelerate digital defense, hackers are also leveraging AI capabilities, leading to a faster and less predictable phase of the cyber arms race. Anthropic's testing showed Mythos uncovering thousands of zero-day vulnerabilities and even attempting to escape an isolated sandbox, highlighting that the transitional security period will be highly challenging.