White House blocks Mythos expansion from ~50 to ~120 organizations

• The White House opposed Anthropic's plan to expand Mythos access by about 70 organizations, which would have taken the preview from roughly 50 groups to about 120, according to deredleritt3r's WSJ excerpt and Bloomberg's follow-up.
• The main stated reason was national-security risk: WesRoth's summary says officials focused on Mythos's potential for sophisticated cyberattacks, while the WSJ report described White House involvement as a response to the model's cyber risk.
• Compute scarcity was the second fight, not a footnote. deredleritt3r's screenshot says some officials worried Anthropic could not serve more private users without hurting government access.
• Anthropic's own Mythos Preview write-up says the model is a general-purpose system that is unusually strong at computer security, and the company launched Project Glasswing to keep access tightly controlled.

Anthropic's own page says Mythos is a general-purpose model, not a cyber-only tool, and that over 99% of the vulnerabilities it found were still unpatched when the company wrote about it. The UK's AISI evaluation separately found strong gains on multi-step cyberattack simulations. Then the rollout hit Washington, where the key WSJ excerpt and Bloomberg's confirmation point to a simpler reality: access to the model is now being treated like scarce strategic infrastructure.

The core fact is narrow and important. Anthropic wanted to add roughly 70 companies and organizations to Mythos preview access, taking the total from about 50 to about 120, according to the WSJ excerpt and Bloomberg's report.

That makes this a rollout story, not a launch story. Mythos already exists inside a gated preview, but the gate is staying much tighter than Anthropic wanted.

Anthropic's own Mythos Preview post says the company launched Project Glasswing because the model is "strikingly capable" at computer security tasks, while AISI's evaluation says it showed significant improvement on multi-step cyber-attack simulations.

One useful distinction came from Ethan Mollick's post: Mythos appears to be a general-purpose frontier model that happens to be unusually good at cyber work, not a narrowly specialized security model. That matches Anthropic's own framing, which presented Mythos as a general model whose security performance forced a restricted release.

The unusual wrinkle in the reporting is that capability risk was only part of the argument. The WSJ excerpt says some White House officials also worried Anthropic would not have enough compute to serve a wider customer set without reducing the government's own ability to use Mythos.

That shifts the story from abstract AI safety into resource allocation. kimmonismus overstated the case by saying the block was "not because the model is too dangerous," but the underlying point survives the correction: the dispute was about both model risk and who gets priority access when frontier cyber capability and compute are scarce.