Not an Emergency
April 12, 2026 · uneasy.in/cca87ab
The Bank of Canada chose the words with care. Its regulators' meeting about Claude Mythos was "situational awareness." Not an emergency.
Five days since Anthropic published its red team assessment of a model it says can discover unpatched vulnerabilities across every major operating system and every major web browser. In that span, Scott Bessent and Jerome Powell called bank CEOs to Washington. The Bank of Canada convened its financial regulators. The Bank of England began arranging its own briefing through the Cross-Market Operational Resilience Group. The EU endorsed Anthropic's staged rollout through Project Glasswing.
Four capitals. No official statement from any of them.
Everything we know comes from anonymous sources speaking to Bloomberg, Reuters, and the Financial Times. The most powerful central banks in the world are treating a language model as a systemic risk to the financial system, and doing it entirely off the record.
Governments have worried about AI before. In committee hearings. In white papers that arrive months after the technology moved on. What I cannot get past is the speed. Washington, London, Ottawa, and Brussels coordinated in days, not quarters. The machinery here is the kind normally reserved for currency contagion.
Gary Marcus argued the response is disproportionate. The demo had sandboxing disabled. Open-weight models can approximate similar analysis. The capability gains track the existing trend line rather than representing a genuine breakout. An AP investigation went further, asking whether Anthropic benefits from the alarm. The company that briefed governments privately before the public knew the model existed, then structured access so that fear becomes exclusivity.
There is something to this. A model too dangerous to release is also a model no competitor can replicate. The IPO preparation is not unrelated. But the skeptics have to explain why Amazon, Microsoft, and CrowdStrike signed up for Glasswing. Either their due diligence failed collectively, or the capabilities are real enough to act on. Forty firms is a lot of firms to fool.
The Bank of England's briefing hasn't happened yet. When it does, it will be closed. No transcript. Whatever they decide will filter out through the same anonymous channels that have carried every detail of this unprecedented week.
Situational awareness. Not an emergency.
Sources:
-
Assessing Claude Mythos Preview's Cybersecurity Capabilities — Anthropic
-
Three Reasons to Think That the Claude Mythos Announcement Was Overblown — Gary Marcus
-
Mythos AI Alarm Bells: Fair Warning or Marketing Hype? — TechXplore / AP
-
UK Financial Regulators Rush to Assess Risks of Anthropic's Latest AI Model — Global Banking & Finance Review
-
Project Glasswing: Securing Critical Software for the AI Era — Anthropic
Recent Entries
- Polo Sport and the All-American Face April 12, 2026
- Mixed Case April 12, 2026
- Three Lines, Cash on Collection April 12, 2026