Overview
- Anthropic, which developed the Mythos model to find and link software flaws, will brief the Financial Stability Board — the G20’s global financial watchdog — on Monday’s reported request from Bank of England Governor Andrew Bailey.
- Anthropic disclosed Monday that Mythos partners can now share vulnerability details and tools with other firms, regulators, and the public under responsible‑disclosure rules, reversing earlier confidentiality limits.
- Access to Mythos remains restricted under Project Glasswing to roughly 40 organizations such as Amazon, Microsoft, Apple, Google, and JPMorgan, with reports that the White House asked Anthropic not to distribute the model more broadly.
- Early users say Mythos boosts bug discovery but also throws false positives and risky fixes, with Cloudflare warning some suggested patches could break dependent code and Palo Alto Networks reporting faster patching alongside error rates.
- Global authorities are preparing next steps as the FSB readies an AI ‘sound practices’ paper for consultation next month and the IMF flags systemic cyber risk, while rivals like OpenAI roll out GPT‑5.5‑Cyber to selected defenders.