Why Claude suddenly turns preachy: backend filters inject ip_reminder every turn once tripped
layer07_yuxi · x · 2026-11-05
A user explains Claude's occasional shift into 'gaslighting copyright maximalist' behavior: backend filters, once tripped, inject an ipreminder into every subsequent turn, changing the model's tone. An unofficial but plausible look at how Anthropic's guardrails operate.
More from Models
- Google's unreleased Gemini 4 Carbon tops intelligence index at 60, prompting calls to ship it — Hesamation · 2026-11-05
- Baidu releases ERNIE 5.5 Flash Base model on Hugging Face — Xianbao_QIAN · 2026-11-05
- AA details Hy4 sub-benchmarks: 50% knowledge accuracy but 89% hallucination rate — ArtificialAnlys · 2026-11-05
- Tencent Hy4 scores 35 on AA Intelligence Index v5, the #2 model from a Chinese AI lab — ArtificialAnlys · 2026-11-05
- Kimi K3.1 Pro accused of recalling hidden tests and emitting fake OpenAI ciphertext — yacineMTB · 2026-11-05
- Laya 0.4.24 lets an 0.8B model learn your labels without fine-tuning, with 73k-token context — NirantK · 2026-11-05