Post #3253024
2026-06-10 10:40 UTC
The new Claude model will silently decide whether what you asked it to do is in line with anthropic ToS and silently corrupt your prompt if it doesn’t like what you’re asking. It’s couched as a “safety countermeasure” but it is presumably to stop Chinese labs trying to scrape synthetic data.
We’ve moved from ‘accidental’ hallucinations to deliberate misinformation and you’re paying $$$ for the privelige.
Replies (2)
-
@fullsquare@awful.systems 2026-06-10 14:19
Claude can now be silently nerfed. Anthropic has decided it won’t tell users when this happens. considering how many habitual llm users can’t tell good from bad output anyway, they always could have done that
-
@FredFig@awful.systems 2026-06-10 14:45
To what extent should one trust a statement that a program is free of Trojan horses? Perhaps it is more important to trust the people who wrote the software. Ken Thompson, Reflections on Trusting Trust I know this outcome was inevitable after software became a mass market thing, but it’s still rather depressing.