AI models such as ChatGPT do not do anything unless a user asks them to.
Mixed Partly true and partly false in roughly equal measure.
Factual: Statements about the present or the past, checked against the record: did they hold up?
What would count as true or false
Judge as of 2025-09-11. True: AI models like ChatGPT took no action that a user had not started or authorised, i.e. every output or action followed a user request or a task the user had set up. Mostly true: this was generally so, with minor exceptions such as user-scheduled tasks or proactive features that a user had to turn on first. Mixed: there were notable products in which the models acted with substantial autonomy beyond the immediate request (e.g. agents taking multi-step actions), so the claim holds only in a narrow sense. Mostly false: widely deployed features had models start actions or messages on their own without user prompting, though they were still broadly user-configured. False: evidence shows that ChatGPT or similar models routinely acted with no user request or setup at all.
Set before the research began.
Evidence (4)
-
E1 Introducing ChatGPT agent | OpenAI OpenAI · 17 July 2025 Own record
“ChatGPT now thinks and acts, proactively choosing from a toolbox of agentic skills to complete tasks for you using its own computer.”
-
E2 Introducing ChatGPT agent | OpenAI OpenAI · 17 July 2025 Own record
“Most importantly, you’re always in control. ChatGPT requests permission before taking actions of consequence, and you can easily interrupt, take over the browser, or stop tasks at any point.”
-
E3 OpenAI Launches ChatGPT Agent That Can Complete Tasks For You MacRumors · 17 July 2025
“OpenAI today launched ChatGPT agent, a new agentic model that is able to think proactively and complete computer-based tasks on the user's behalf.”
-
E4 Introducing ChatGPT agent | OpenAI OpenAI · 17 July 2025 Own record
“Additionally, you can schedule completed tasks to recur automatically, such as generating a weekly metrics report every Monday morning.”
Reasoning
An independent report (E3) shows that ChatGPT agent launched in July 2025 as an agentic model that "think[s] proactively" and carries out computer-based tasks for the user. OpenAI's launch page (E1) says it proactively chooses its own tools and steps, using its own computer. The page also says users stay in control, the agent asks permission before consequential actions, and users can schedule completed tasks to recur automatically (E2, E4). So by the claim date, a notable product proactively chose from a toolbox of agentic skills to complete computer-based tasks on a user's behalf, though a user started each task and it asked permission before consequential actions (E1, E2, E3). Under the criteria, that means the claim holds only in a narrow sense.
Cite this claim
Sam Altman, “AI models such as ChatGPT do not do anything unless a user asks them to.” (said 11 September 2025). Verdict: mixed. Reliability API, claim c_e0dc850e526f, method 0.1.0+c876030c, dataset release v1.1.2. https://epistacklabs.com/claims/c_e0dc850e526f
This claim is in dataset release v1.1.2, with the same verdict.
Reply to this claim
More factual claims by Sam Altman
-
Mostly true
“not until 2020 did we train GPT-3, which was the first model that I think actually was over some threshold of usability.”
OpenAI trained GPT-3 in 2020, which Altman considers the first model above a threshold of usability.
-
Resolved true
“that it is so capable that it is discovering new knowledge, doing science, able to add huge economic value, create entire pieces of complex software.”
-
Mixed
“They now hallucinate a little bit.”
-
Mostly true
“But in terms of figuring out what questions to ask or to work on something over a very long period of time, we are definitely not close to human performance.”
-
Mostly true
“It’s an integrated model, so you don’t have to pick in our model switcher and know if you should use GPT-4o or o3 or o4 mini or any of the complicated things.”
Verdicts are made automatically; how verdicts are made.