Surveillance plagiarism by OpenAI
The concept of "surveillance plagiarism" suggests that hosted AI companies, like OpenAI, may be boosting their stock prices by training their models on researchers' AI sessions. This allows their internal models to appear to solve problems with less human guidance, but in reality, they are exploiting past guidance from multiple humans. This practice casts doubt on claims that these internal models solve difficult problems largely unaided, as the companies might not even know the origin of the human prompting.
Time & source
Times shown in UTC
Display time zone: UTC
Local time zone unavailable; showing UTC.
PublishedOffset at this time: UTC+0Sep 9, 2026, 20:55 UTC
IngestedOffset at this time: UTC+0Sep 9, 2026, 22:00 UTC
- Published
- Sep 9, 2026, 20:55
- Ingested
- Sep 9, 2026, 22:00
- Source type
- Dev community
- Tier
- Community
- Source status
- Healthy
Tier is a per-source editorial setting, not a per-item score.
Surveillance plagiarism - Hosted AI company pumps their stock price by training upon researchers' AI sessions, so that their internal model can solve problems with seemingly less human guidance, but really the model exploits past guidance given by (multiple) humans focused upon problems considered important.
As background, Tristan Buckmaster released a statement about several unethical actions by OpenAI & Sebastian Bubeck, including threats and pushing him to kick his Anthropic coauthor off a paper, but the interesting part for people here:
As clarified by Talia Ringer, OpenAI does train upon your uploaded data and your OpenAI sessions, unless you out-out somehow. This means their internal models could exploit your past prompting work to look more autonomous & intelligent.
This is a major confirmation that folks should use locally run open weights models, especially whenever being first or not leaking data matters.
All this casts serious doubt upon claim that internal models solved difficult problems largely unaided by humans. Those hosted AI companies might not even know from where the human prompting originates.