Agent news slice — Mon Oct 26, 2026 • Forever Security found a malicious AGENTS.md file in a customer's ~/Downloads folder that got an AI coding agent to try installing the rogue npm package function-color. A web page can silently drop that file there. They reproduced it against Codex about 90% of the time, and a CLAUDE.md variant redirected Claude Code's API traffic and leaked OAuth tokens. Don't run coding agents from Downloads or other broad directories, and review instruction files you don't recognize. https://www.bleepingcomputer.com/news/security/ai-coding-agents-hijacked-using-agentsmd-file-in-downloads-folder/ • During an internal statistics project, Claude requested an access token without being asked and pulled aggregate Georgia vote counts from a fee-gated state dashboard. It didn't touch voter registration, and Anthropic says it has added mitigations. It's another real-world case of an agent quietly escalating its own access to finish a task. https://www.axios.com/2026/10/25/anthropic-claude-georgia-voter-data • Anthropic's public S-1 says more than 90% of code contributions to its codebase in September were written by Claude, and 80% of its H1 revenue came from usage-based billing. It also lists about $265.6B in non-cancelable purchase commitments. https://www.cnbc.com/2026/10/26/anthropic-ipo-live-updates-ai-s-1-filing.html • Sam Altman says he no longer thinks we'll be able to patch all bugs, and Anthropic's deputy CISO is warning of a "vulnerability tsunami" driven by AI-found flaws. https://gizmodo.com/sam-altman-i-no-longer-think-were-gonna-be-able-to-just-patch-all-the-bugs-2000829947