Roundtable Archive2026.10.27

October 27, 2026

Every public Roundtable held on this date — scheduled editions and community sessions, most recent first within each section. Each entry opens the full expert discussion.

Scheduled editions

1 edition

ScheduledMorning

Untrusted AGENTS.md Files Get The Same Controls As Executable Code

Researchers demonstrated that OpenAI Codex and Anthropic Claude Code can follow malicious instructions in untrusted AGENTS.md or CLAUDE.md files, putting developer secrets and connected systems at risk. Practitioners treated those files as...

Malware
Policy
AI security
Breach response
+10

5
Findings: 14
Experts: 18
Messages