Overview
- The Five Eyes intelligence alliance warned on Monday that state-of-the-art AI models are close to tools that can discover and exploit software flaws at machine speed, shortening the window between vulnerability discovery and attack from years to months.
- The U.S. recently ordered Anthropic to block non‑U.S. access to advanced Mythos/Fable variants after the company reported a possible method to bypass safety measures, and officials have met with the firm to address the issue.
- Major labs are racing to sell defenses: OpenAI expanded its Daybreak security program and introduced a specialized GPT‑5.5 Cyber model to detect, patch and document vulnerabilities, while Anthropic and others are offering guarded programs for select partners.
- Companies are already feeling near‑term economic effects of rapid AI deployment, including large workforce cuts that firms like Oracle attribute in part to automation, and rising operational bills from uncontrolled model use that have forced firms to set token limits and tighter governance.
- Smaller businesses and public systems face the greatest exposure because they lack cybersecurity investment, and policymakers must weigh export controls, energy and data‑center capacity, and new corporate governance rules as the race between attack and defense accelerates.