BoE to Brief on 'Mythos'

The Bank of England plans to brief financial firms on Anthropic’s new Mythos model after officials judged its cyber capabilities relevant to banks, insurers and market plumbing. (bloomberg.com) Bloomberg reported on April 11 that the central bank would discuss Mythos with financial institutions as United Kingdom regulators joined counterparts in the United States in raising alarms about the tool. Reuters reported on April 12 that British financial regulators were already in urgent talks with the National Cyber Security Centre and major banks. (bloomberg.com) (msn.com) Mythos is a large language model, a system that predicts and writes the next likely token in a sequence, but Anthropic says this version is unusually strong at software security work. In its system card, the company said Mythos Preview was its most capable frontier model to date and showed a sharp jump over Claude Opus 4.6 on many benchmarks. (anthropic.com) Anthropic’s red-team researchers said Mythos could identify and then exploit zero-day vulnerabilities in every major operating system and every major web browser during internal testing. A zero-day is a previously unknown software flaw, the kind of bug defenders have not patched yet because they do not know it exists. (red.anthropic.com) That matters for central banks because modern finance runs on software: banks, exchanges, payment systems and cloud providers all depend on code that can fail or be attacked. Reuters said United States Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell called bank chief executives to Washington on April 9 for a closed-door discussion of the same risk. (msn.com) Anthropic has not opened Mythos broadly. The company said it first deployed the model internally, then released it only to a small set of external users through a limited research access program. (anthropic.com) The company also created Project Glasswing, a program that gives selected defenders access to Mythos to harden critical software before wider misuse becomes easier. Anthropic said the initiative includes launch partners using the model for defensive security work and more than 40 additional organizations that build or maintain critical software infrastructure. (anthropic.com) Some officials are treating the model as a direct financial-stability issue, not only a technology issue. Reuters said British regulators were assessing whether Mythos’s ability to find hidden flaws could become a risk for banks, insurers, exchanges and payment companies. (msn.com) Anthropic’s own risk report took a narrower line on another fear around advanced systems: autonomy. The company said Mythos appeared to be its best-aligned released model so far, even as it warned that the model was significantly more capable and more agentic than prior versions. (anthropic.com) The next step is not a public ban or a broad launch, but more closed-door risk reviews. For now, the Bank of England is treating Mythos less like a new chatbot and more like a new source of cyber exposure for the financial system. (bloomberg.com)