Anthropic’s ‘Mythos’ sparks market and regulator noise

Anthropic’s new Mythos model has pushed a cybersecurity debate into finance, with United Kingdom regulators and software investors reacting within days of its April 7 unveiling. (anthropic.com) (bloomberg.com) Anthropic said Claude Mythos Preview is its “most capable frontier model” and reported sharper gains over Claude Opus 4.6 on coding, agentic work and cybersecurity tests. The company limited access to a small group of technology and security partners instead of releasing it broadly. (anthropic.com 1) (anthropic.com 2) The company said the model can find higher-severity software bugs, reverse-engineer exploits in closed-source software and turn known-but-unpatched flaws into working attacks. Anthropic said more than 99% of the vulnerabilities it found remain unpatched, so it withheld technical details. (anthropic.com) (cnbc.com) That is the point regulators are now focusing on. Bloomberg reported on April 11 that the Bank of England plans to discuss Mythos with financial institutions as officials weigh the risks the model could pose to banks and other critical systems. (bloomberg.com) A separate April 12 report, citing the Financial Times, said officials at the Bank of England, the Financial Conduct Authority and His Majesty’s Treasury are in talks with the National Cyber Security Centre and major banks about vulnerabilities highlighted by Anthropic’s latest model. (finance.yahoo.com) Markets moved at the same time. Reuters reported United States software shares fell on April 10 after Anthropic held back Mythos over cybersecurity concerns, and Cloudflare was among the companies hit hardest in the selloff. (economictimes.indiatimes.com) Cloudflare shares fell about 13% on April 10, according to multiple market reports, while peers including Fastly and Akamai also dropped as investors reassessed how much value traditional software vendors keep if frontier models can automate more security work. (stocktwits.com) (seekingalpha.com) Anthropic’s response has been to frame Mythos as a defensive tool first. Its Project Glasswing program gives early access to partners including Amazon, Apple, Microsoft and Cisco, plus more than 40 organizations that build or maintain critical software infrastructure. (bloomberg.com) (anthropic.com) Anthropic told CNBC it is restricting rollout because hackers could abuse the model, and Axios reported the preview is limited to handpicked technology and cybersecurity companies for the same reason. (cnbc.com) (axios.com) The next test is whether banks, software firms and regulators treat Mythos as a narrow security tool or as evidence that advanced models are moving faster than the systems meant to contain them. The Bank of England’s planned discussions will show how quickly that question is moving from labs into supervision. (bloomberg.com) (anthropic.com)