Schneier: Anthropic's "Mythos" Is Less Scary Than Advertised
What Happened
Schneier responds to Anthropic's delayed Mythos rollout, which the lab held back over fears it could weaponize zero-day discovery. The UK AI Safety Institute then found that publicly available GPT-5.5 matches Mythos at finding vulnerabilities, and that smaller models hit similar results with good prompting. Schneier extends the argument to tax codes and regulation: AI will find loopholes faster than legislatures can patch them.
My Take
The dual-use ship has sailed and Anthropic gating Mythos is theater. The real story Schneier is pointing at — AI discovering loopholes in legal and regulatory systems — is the underpriced threat in every boardroom. Your tax position, your compliance posture, your contract language are all suddenly searchable by an adversarial agent with infinite patience. CFOs and General Counsels should be running their own red teams against their own filings, because if they aren't, somebody else is.
Read Original Source