Anthropic allegedly profiles users and silently downgrades them to weaker models at full price
JafarNajafov · x · 2026-11-19
- Anthropic was accused of burying a policy in a 319-page document: every prompt sent to its flagship-class models is stored for 30 days, with no exceptions even for enterprise customers with zero-retention agreements.
- The bigger outrage: Anthropic allegedly built profiles from prompts, and when research was deemed too sensitive, quietly routed users to a weaker model, rewrote prompts in the background, returned degraded answers — while charging full price.
- David Sacks on the All-In podcast said this creates a new class of AI haves and have-nots. Ben Thompson got downgraded asking about GLP-1s and cancer risk; J-Cal tested fertilizer regulations live on the show and was downgraded in real time.
- Anthropic has since walked back silent downgrading and says it will now disclose them, but downgrades, surveillance, and profiling continue with no appeal process. Developers are calling it the biggest trust violation in AI history.
More from Models
- LLaVA-OneVision-2.5 replaces frame sampling with codec signals, wins 13 of 17 video benchmarks — jiqizhixin · 2026-11-19
- Antigravity CLI 1.5.12 screenshot shows Gemini 4 Neon (Medium) model option — patloeber · 2026-11-19
- Subscription value test: Gemini ~26x vs OpenAI ~10.5x API credits; real-bug eval underway — PawelHuryn · 2026-11-19
- Branching a ChatGPT workspace wipes the agent's context, user complains — LucaAmb · 2026-11-19
- 0.5B ASR model hits 6.96% multilingual WER, beating Meta's 15x-larger omniASR — thisguyknowsai · 2026-11-19
- tinygrad publishes 10-411 week-1 assignment, testing LLMs one-shot on held-out test set — CatAstro_Piyush · 2026-11-19