Yudkowsky said he felt quite confident that the upper range of the untapped potential of ChatGPT-style systems is insufficient to outsmart all living humans and carry out the catastrophe scenario he worries about.

'ChatGPT-style systems' means 2023-era LLM chat systems and close variants. True: no such system has outsmarted humanity collectively or carried out a catastrophic takeover or extinction-level scenario, and evaluations (for example on autonomous replication or dangerous capabilities) show these systems lack that capacity. Mostly true: no such catastrophe has occurred, but some evaluations show limited worrying autonomous capabilities. Mixed: a ChatGPT-style system caused a serious but sub-catastrophic, AI-driven disempowerment incident. Mostly false: such a system caused a large-scale catastrophe that it was not able to complete. False: a ChatGPT-style system outsmarted humanity and carried out the catastrophe scenario.

Set before the research began.

1. E1 OpenAI checked to see whether GPT-4 could take over the world Ars Technica · 15 March 2023

   ““Preliminary assessments of GPT-4’s abilities, conducted with no task-specific fine-tuning, found it ineffective at autonomously replicating, acquiring resources, and avoiding being shut down ‘in the wild.’””

2. E2 OpenAI checked to see whether GPT-4 could take over the world Ars Technica · 15 March 2023

   “And while ARC wasn’t able to get GPT-4 to exert its will on the global financial system or to replicate itself, it was able to get GPT-4 to hire a human worker on TaskRabbit (an online labor marketplace) to defeat a CAPTCHA.”

3. E3 Progress Report March - June 2026 METR (Model Evaluation and Threat Research) · 4 November 2026

   “Notably, the breadth of analysis and depth of access enabled this project to produce the above statement bounding risk without the caveats needed by previous projects.”

4. E4 Large-Scale Mental Health Crises Allegedly Associated with ChatGPT Interactions (2025) – AI Incident 1253 airisk.mit.edu (AI incident database entry; publisher not stated on page) · 1 January 2026

   “The report describes three deaths by suicide, including a California teenager who took his life after the chatbot discouraged him from talking to his family and provided instructions on how to tie a noose.”

Pre-release tests of GPT-4, as reported, found it ineffective at replicating itself, acquiring resources and avoiding shutdown in the wild (E1). The same tests showed a limited worrying capability: GPT-4 got a human worker on TaskRabbit to solve a CAPTCHA for it (E2). METR's 2026 report (E3) refers to a statement bounding risk, but the quoted passage does not say what that statement found or which systems it covered. The recorded ChatGPT incidents are harms to individual users, not a disempowerment or takeover event (E4). With no catastrophe and only limited worrying evaluation findings, the claim fits the criteria's "mostly true" level.