Microsoft reported that malicious LiteLLM 1.107.0 packages reached PyPI and official container images on October 21. The payload could run when Python started, without an application importing LiteLLM, and targeted credentials from developer tools, cloud environments, browsers and AI systems.
Anthropic donated Agent Skills, its open format for packaging agent instructions and workflows, to the Linux Foundation’s Agentic AI Foundation. More than 70 agent platforms already support the standard, which will now use shared maintainers and a public proposal process.
Google introduced Gemini 4 Carbon, a frontier model for finding, verifying and patching software vulnerabilities. Access is limited to selected cyber defenders through Fairwind, making the launch a controlled deployment of long-horizon security capabilities rather than a general release.
