AI News Nuggets

Agent settings and vulnerability intake need named owners

Google reports intrusions that used coding-agent configuration as a malware path. Anthropic has announced shorter disclosure windows for its human-reviewed reports from 2027, while GitHub has added a control over where Copilot customizations originate. The common operating question is who can change agent behavior, receive findings, and approve extension sources.

Editorial read

This edition collects 3 notes across 3 topic areas and 3 sources. Start with Review coding-agent settings as executable configuration, Plan for shorter disclosure windows on validated AI-found flaws, Constrain which sources can extend coding agents to get the week's main practical signal before scanning the remaining links.

Edition signal

The November 17 signal is to review the paths that carry agent authority

Pick one coding-agent workflow. Check who can edit its settings and hooks, which plugin or MCP sources it loads, and whether changes require review. Then trace one confirmed vulnerability report from intake to patch, disclosure decision, and verification. Assign owners and test these paths before widening use.

SecurityGovernanceTools
Governance
Official policy update

Plan for shorter disclosure windows on validated AI-found flaws

Source: Anthropic

Anthropic says its coordinated disclosure window will be 30 days for high- and critical-severity issues and 45 days for others when it reports to vendors or maintainers on or after January 1, 2027. It plans to seek an external coordinator after 15 days without a response. The seven-day window for active exploitation remains; the new deadlines do not apply to unreviewed OSS Scanner findings. This is Anthropic's policy, not a universal disclosure rule.

Why this matters: If you maintain software, give security reports a monitored intake, a reproducibility owner, and a path from confirmed impact to patch and regression check. Agree on realistic remediation and disclosure milestones for each report, including hard-to-deploy fixes, before a shortened window becomes a surprise.

Read Anthropic's disclosure policy update
Tools
Official changelog

Constrain which sources can extend coding agents

Source: GitHub

GitHub says Enterprise Cloud administrators can require selected Copilot customizations to come from plugins or managed settings rather than standalone user or project configuration. The categories include skills, custom agents, hooks, MCP servers, and executable extensions. The control applies to supported Business and Enterprise users; plugin-only by itself does not restrict which marketplaces supply those plugins.

Why this matters: Choose which extension categories need central control and test the effective policy in a real client. Pair strictPluginOnlyCustomization with strictKnownMarketplaces and specific plugin rules where an approved source list is required. Review repository defaults and exceptions before calling the rollout governed.

Read GitHub's Copilot customization update