Claude Opus 4.8 processed the identical query without issue.
Third request to Fable and it triggered - a paper on panspermia ethics looked dangerous because it mentions bacterial spores. Yes, Fable seems to be prone to panicking. (Opus 4.8 plods along, and blithely suggests expanding the paper in all sorts of cool directions.)
Users criticize Fable AI for over-refusing biology prompts like taxonomy visualizations or panspermia ethics papers, viewing the blocks as unfounded and overly cautious.
@anderssandberg Yes, it's pretty clearly over-refusing:
@anderssandberg And they are aware of the problem, but don't have a great solution.
To be fair, there is a fundamental false positive/false negative tradeoff - and the false negatives get exploited, so for dangerous enough capabilities, that rate needs to be zero.
@anderssandberg Fable won't respond to any of my prompts unfortunately, seems like being interested in biology is being treated as a red flag for anything else :p
@aaarataki Yes, it has serious biologyphobia. "Could you come up with a good way of doing a visualisation of the taxonomy of the coleoptera?" triggered a safety warning.
@aaarataki The plausible explanation is that Anthropic has just failed at instilling discernment in the model. The fun/creepy/unlikely explanation is that biology actually is riddled with serious risk, but we are not noticing it...
Claude Opus 4.8 processed the identical query without issue.
Third request to Fable and it triggered - a paper on panspermia ethics looked dangerous because it mentions bacterial spores. Yes, Fable seems to be prone to panicking. (Opus 4.8 plods along, and blithely suggests expanding the paper in all sorts of cool directions.)