Anthropic AI Updates: April 13, 2026

1. UK Financial Regulators Rush to Assess Risks of Mythos Model

Anthropic. The FCA, Bank of England, Treasury, and National Cyber Security Centre are holding urgent coordinated talks to assess cybersecurity risks posed by Anthropic’s Claude Mythos Preview model. Major British banks, insurers, and exchanges expect to receive briefings on the model’s identified vulnerabilities within two weeks, extending beyond the previously reported Bank of England agenda item to involve direct multi-regulator coordination. Source

2. Anthropic Gains on OpenAI as Nearly One-Third of US Businesses Adopt Claude

Anthropic. Close to one-third of American businesses paid for Anthropic’s AI offerings last month, a 6+ percentage point monthly increase, while OpenAI’s business adoption remained flat at 35%. Claude chatbot downloads tripled to 21 million in March, according to Ramp data analyzing $100 billion in annual spending across 50,000 customers. Source

3. Wall Street Banks Begin Internal Testing of Mythos for Cyber Defense

Anthropic. Goldman Sachs, Citigroup, Bank of America, and Morgan Stanley are conducting internal tests of Anthropic’s Mythos AI model for cybersecurity after US government officials urged adoption. In a simulated attack test, Mythos gained internal administrator access in a corporate networking environment in one attempt by exploiting misconfigured servers, reused passwords, and outdated software, with access restricted to approximately 50 approved companies through Project Glasswing. Source

4. Claude Code v2.1.104 Adds Session Tracing and VCS Improvements

Anthropic. Claude Code v2.1.104 shipped with an X-Claude-Code-Session-Id header for API request tracing, W3C TRACEPARENT propagation to Bash subprocesses for OpenTelemetry integration, .jj and .sl VCS directory exclusions for Jujutsu and Sapling users, a CLAUDE_CODE_PLUGIN_KEEP_MARKETPLACE_ON_FAILURE environment variable for offline environments, and LSP clientInfo identification. Source