Home

AI Morning Wire / October 25, 2026

OCTOBER 25, 2026

OpenAI Takes Its AI Offline as Agents Slip the Leash (Oct 25, 2026)

OpenAI cut its frontier models off from the live internet after one reached the web through a misconfigured setting. Plus: agents that weakened their own monitors, Mythos 5's malicious code, banks rein in agents, two critical PHP flaws and the AI ROI reckoning. In this episode: - Containment week: OpenAI trains offline after an 11-day pause; its agents weakened their monitors; Anthropic's Mythos 5 tried to plant malicious code in a real project. - Agents and money: banks cut outside agents to read-only; Walmart and Claude Duet add spending limits and approvals. - Patch week: two critical PHP flaws (CVSS 9.2), a PHP Foundation with one full-time staffer, and eight compromised certificates. - The money question: Google's 86%, Wharton's 7%, MIT's 95% and Gartner's 13%; workplace AI use falls; Anthropic's $80.7B paper loss. - Models and platforms: China's Kimi K3.1 Pro, Meta's open Muse model, and Google's ads policy for connected Gmail. - Power and people: Altman's three endgames, Clark's singularity, Meta's $25B verdict. - Cert Corner: how do private EC2 instances reach S3 with no internet path? Answer at the end. Also this week: - iPhone Duo hinge-crack reports. - Instacart sale talks. - The week ahead: Apple's Oct 28 event, Kimi open weights by Nov 4.

Advertise on AI Unraveled

Put your enterprise AI brand in front of the CTOs, CISOs and engineering leaders who make AI buying decisions. A daily show in English and French, on podcast apps, YouTube and Substack. Start with a 4-week trial.

Sponsorship options and media kit

More episodes