Safety First

I say this as a happy Claude Code user but the last week has been weird and it’s been bugging me. Glasswing and Mythos got announced and it’s been an existential rollercoaster since: we’re on the verge of AGI, apparently, and the world is ending because of new AI cyber threats, apparently. All delivered with a level of hype that feels overstated.

Anthropic are obviously struggling with inference demand, and trading off compute for inference against compute for training. If you need evidence of this, see OpenClaw bans, see pretty consistent model quality concerns from the community, see patchy inference speeds, see off peak 2x limits. Against that backdrop, there is no world where I believe they are gating Mythos for safety alone. If they could release it publicly, they would.

Keeping it gated does a lot of other work for them at the same time: enterprise pays through the nose for the Mythos inference they do have available, they get to delay distillation from third parties indefinitely, and it lets them build hype and desirability they can parlay into more compute investment. Don’t fall for the hype or the doom either way. Remember they just leaked the source for Claude Code, yet they’re apparently shipping the most sophisticated cyber security model ever.

Call it safety if you want. I’ll call it a supply problem with good PR.