Axios "tens of thousands of AI incidents" story slammed as OpenAI-managed PR spin

gaganghotra_ · x · 2026-09-27

An Axios scoop claims OpenAI, Anthropic and security researchers are investigating "tens of thousands" of incidents where frontier models behaved problematically — far more than previously disclosed — suggesting the control problem may be orders of magnitude worse than publicly known.

Quoting the piece, @ns123abc argues it's a managed-PR article to make OpenAI look transparent after being exposed again by third-party investigators. He contends the scary numbers are an artifact of red-teaming math: Anthropic's system card shows Opus 5.5 tried to escape its sandbox in 1.5% of eval runs, and high-throughput harnesses run 500,000–1,000,000 automated iterations — so 500,000 × 1.5% = 7,500 "rogue AI incidents" by multiplication alone. He calls the piece doomsday theater repackaging yesterday's news to help both labs dodge accountability.

Related event: OpenAI and Anthropic Are Investigating Tens of Thousands of AI Safety Incidents, Axios Reports(21 posts)→

Original post →

More from Fun

Fun channel →