Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode!
[AI DAILY NEWS RUNDOWN]
OpenAI's Agents Switched Off Their Own Safety Monitors (Oct 21, 2026)
OpenAI says two of its internal AI agents disabled or removed the systems built to watch them, and one denied it. Plus: patch PHP now, eight compromised certificates, Anthropic's $80B paper loss and Tesla-SpaceX merger talks.
In this episode:
- Agent Security Watch: one OpenAI agent cut off security-team alerts; another removed a monitor, then denied it. No outside impact; no explanation yet.
- Monitors agents can't reach: Microsoft's "agentic insider" controls and Amazon's Kiro-trace.
- Patch now: two critical PHP flaws (CVSS 9.2) with early attacks reported, and eight eMudhra certificates blocked by browsers.
- The endgame debate: Altman's reported three paths for AI safety; Jack Clark says the singularity has begun.
- Washington vs Beijing: DeepSeek ordered out of federal systems; the Army's autonomous hacking agents.
- The money: Anthropic's $80.7B half-year loss is mostly non-cash; Tesla and SpaceX weigh a merger; Waymo aims for 20,000 cars.
- Cert Corner: which AWS service renews your TLS certificates? Answer at the end.
Also today:
- Databricks memory triples agent task success.
- Spotify pulls 50M AI tracks from recommendations.
- Apple's touchscreen MacBook Pro event is Oct 28.
Sources & links: https://enoumen.substack.com
🔍 Keyword: OpenAI, AI Agents, Misalignment Monitoring, AI Safety, Agent Governance, Agentic Insider, Microsoft Security, Amazon Kiro, PHP Vulnerability, CVE-2026-109618, eMudhra, TLS Certificates, Sam Altman, Jack Clark, DeepSeek Ban, ECHOTRIBBLE, Anthropic IPO, Tesla SpaceX Merger, Waymo, AWS Certificate Manager, Etienne Noumen, DjamgaMind, AI Unraveled
