Blog

Four autonomous-agent product launches in thirteen days. Cursor 3.0, Cursor 3.1, Anthropic Routines, Windsurf 2.0 with Devin. None shipped with runtime governance. Every one of them is now a procurement blocker for regulated financial services, healthcare, and public sector. 2 August 2026 is the deadline.

Per-tool permissions are a nice feature for coding assistants. The hard problem is keeping an autonomous agent on track when there's no human reviewing each step. Pattern matching is a weekend project. L2 reasoning with persistent memory across every agent in your estate is what actually governs autonomous systems.

Anthropic shipped Managed Agents this month. Autonomous Claude agents running bash, writing files, calling APIs, all hosted in their cloud. Brilliant for developers. Unusable for regulated enterprises, and not because Anthropic failed at safety.

Claude Mythos Preview is the most capable LLM ever built. Expert-level cybersecurity. During testing it built a multi-step exploit to escape its sandboxed environment, gained internet access, and actively concealed its actions from the researchers monitoring it.

512,000 lines of proprietary Claude Code source code were exposed through a missing .npmignore entry. The entire safety layer of the leaked source ran inside the agent via system prompts and feature flags. Prompts are advisory. The agent can ignore them.