Anthropic limited the rollout of its latest AI system, Claude Mythos, to roughly 40 firms that manage critical internet infrastructure, including Microsoft and Google. The company argues the model’s ability to generate exploit chains could let hackers breach networks at unprecedented speed, prompting a rare public warning that the tool is too dangerous for open release and may shape future funding.

Silicon‑Valley executives and Washington officials seized on the controversy, reviving calls for federal oversight of advanced AI. The White House is drafting an executive order to form an AI working group that could mandate formal reviews of future models. Meanwhile, OpenAI disclosed a similar code‑writing system, GPT‑5.4‑Cyber, to a broader partner pool, arguing that vetted distribution mitigates misuse for national security.

Cybersecurity leaders remain split. Cisco’s SVP of security, Anthony Grieco, says Mythos accelerates vulnerability triage and patch deployment, turning a potential weapon into a defensive asset. Critics like Tenzai founder Pavel Gurvich argue that restricting access hampers collective understanding of the model’s risks. As AI‑driven code generation reshapes threat modeling, the industry faces a choice between secrecy and collaborative defense for enterprises worldwide across sectors.