Claude Code 2.1.325 ships significant security improvements: sandboxed Bash commands now require permission by default, a zstd decoding vulnerability is patched, and sandbox processes can no longer overwrite Claude Code's own settings or access API keys. Here's why these changes matter.
Anthropic opened the Claude Security public beta to Enterprise customers on October 21, 2026. Rather than scanning once and moving on, it monitors repositories continuously, re-analyzes changes as they happen, and tracks each vulnerability until it is fixed. Team-informed threat models and API/MCP integrations are included.
Claude Code v2.1.283 (September 25, 2026) strengthens team-level AI model management. Admins can now block specific models with deniedModels and prevent automatic adoption of new releases with availableModelsMatch: "exact". Here's what every administrator of an enterprise or educational institution using Claude Code needs to know.
Google unveiled Gemini 3.8 Flash Cyber on September 2, 2026. It achieves 70%+ success on autonomous vulnerability discovery across 20 programming languages and delivers 2.6Γ more accurate Chrome bug patches than competing models. Tasks that once took months now complete in under 2 hours. Access is limited to defenders through Google's Fairwind Program.
Gemini Spark now uses your real Chrome browser and saved logins to run web errands on your behalf. Anthropic launched Inference Hooks for Claude Enterprise, inspecting every employee prompt before it reaches the model. Suno announced mandatory AI music watermarking and download limits amid ongoing copyright lawsuits.