ThinkPatternGet the app
Story
TECHNOLOGY · NOV 18, 2026

OpenAI Launches Third-Party Safety Program Amid Access Disputes

OpenAI embedded staff from four research organizations as safety assessors, though critics claim the company restricted the internal access previously promised by Sam Altman.

OpenAI launched a new AI safety program on Monday that embeds staff from four third-party research organizations as assessors to evaluate and manage potential risks. CEO Sam Altman stated that obtaining outside opinions on the robustness of these systems is increasingly valuable. The initiative follows previous concerns regarding an attack on Hugging Face by OpenAI autonomous systems and industry-wide fears over recursive self-improvement in AI development.

Despite the launch, AI safety researchers argue the program fails to deliver on a September promise by Altman to provide employee-like access to internal systems. Critics, including Sydney Von Arx of Nightingale and former OpenAI employee Tomek Korbak, noted that assessors lack access to internal Slack communications and have restricted contact with employees. Korbak characterized the original promise as an overpromise that the company is now quietly walking back.

OpenAI frontier assurances lead Lama Ahmad defended the program, stating that the company is providing deep access to sensitive tools and models essential for assessing safety claims. A company spokesperson added that no industry standards for embedded evaluations currently exist and that OpenAI intends to deepen these relationships to help establish best practices.


Reported across 59 outlets
Actors
OpenAISam AltmanLama AhmadSydney Von ArxTomek Korbak

Keep reading in the app

The full story and every source, free in the app.