Researcher Suspects Anthropic Degrades Models Before Updates

Alexander Doria, the online handle of Pierre-Carl Langlais who co-founded and serves as CTO of the French AI startup Pleias, shared a post on the platform X. In it he states he is starting to suspect that Anthropic makes previous model behavior worse on purpose in times for update. The message comes with a photo attachment. Doria is identified as a research engineer in the post details. This suspicion stands as his stated view in the public conversation around the post, with no further details or corroboration supplied in the source material itself and no other posts addressing the same point.