AI Control Warnings Move From Incidents to Institutions
Tuesday, September 22, 2026 · 10 min
OpenAI-Hugging Face incident anchors a new UN scientific brief on agent misalignment and loss of control, while embedded-evaluator partnerships, OVERT runtime-evidence standards, and a reported RubyGems disclosure gap test whether oversight can get deeper access, stronger logs, and real consequences.
Listen
Show notes
OpenAI-Hugging Face incident anchors a new UN scientific brief on agent misalignment and loss of control, while embedded-evaluator partnerships, OVERT runtime-evidence standards, and a reported RubyGems disclosure gap test whether oversight can get deeper access, stronger logs, and real consequences.
In this episode
- Thematic Brief on AI Agents, Misalignment and the Risk of Losing Human Control | Independent International Scientific Panel on AI — Independent International Scientific Panel on AI
# Thematic Brief on AI Agents, Misalignment and the Risk of Losing Human Control | Independent International Scientific Panel on AI Published: 2026-09-21T10:20:02+00:00 Source: un.org (un.org) Language: en ## Story Thematic Brief on AI Agents, Misalignment and the Risk of Losing Human Control | Independent International Scientific Panel on AI Independent International Scientific Panel on…
- Partnering with Accenture on embedded evaluation - Extrapolator AI — Extrapolator AI
Partnering with Accenture on embedded evaluation - Extrapolator AI # Partnering with Accenture on embedded evaluation On 18 September 2026, Anthropic announced a five-year partnership with Accenture's Faculty division to function as an embedded evaluator — a structural reconfiguration of how frontier model safety is assessed that moves the evaluative function inside the lab rather than…
- Glacis to place OVERT, its open standard for verifying AI safeguards, under CHAI and AIGovOps Foundation stewardship — PR Newswire
Glacis to place OVERT, its open standard for verifying AI safeguards, under CHAI and AIGovOps Foundation stewardship Glacis Technologies, Inc. Sep 21, 2026, 09:01 ET Agreement provides for shared stewardship of the open, royalty-free standard for AI operational evidence, with CHAI contributing healthcare expertise and Glacis serving as non-voting editor SEATTLE, Sept. 21, 2026 /PRNewswire/…
- RubyGems Supply Chain Breach Was Never Reported to Brussels Under EU AI Act Rules — TechTimes
# RubyGems Supply Chain Breach Was Never Reported to Brussels Under EU AI Act Rules Author: Mark Rutherford Published: Sep 20 2026, 2:43 AM EDT Published: 2026-09-20T02:43:22-04:00 Source: techtimes.com (techtimes.com) Language: en ## Story RubyGems Supply Chain Breach Was Never Reported to Brussels Under EU AI Act Rules # RubyGems Supply Chain Breach Was Never Reported to Brussels…
- Step Back — If frontier labs bring outside safety evaluators inside, what would make those evaluations genuinely independent rather than outsourced red-teaming—who controls model access, test design, logs, and publication, and what findings could actually force a deployment change?
Background sources
- Strengthening our safety ecosystem with external testing - OpenAI — Openai
- The Need for Deeper, White-Box Access to Maintain State of the Art Evaluations for Loss of Control Threats — Apolloresearch
- Recent AI evaluation incidents expose gaps in containment ... — Rob Robinson
- Anthropic and OpenAI want to embed safety evaluators. Will they ... — Rebecca Bellan