User Challenges Anthropic: Is Opus 5 Really More Aligned Than Previous Models?
dhadfieldmenell · x · 2026-08-19
Anthropic claims Opus 5 is their most aligned model based on automated behavioral audits, showing the lowest rates of reckless or deceptive behavior. However, user EzraJNewman questions this, stating that Opus 5 seems clearly less aligned than previous Claudes, suggesting the automated audit metrics may not reflect real-world alignment or user experience accurately.
More from Models
- Feldar's longform AI models bypass AI detection tools — almmaasoglu · 2026-08-19
- User test: DeepSeek V4 Flash beats GPT 5.6 in coding — AnuranBuilds · 2026-08-19
- After a cyberattack on France's tax authority, fears Mistral's non-frontier models will guard critical systems — eliebakouch · 2026-08-19
- Reddit hails Qwen 3.8 27B as 'DeepSeek moment' for local models — InternationalGap3698 · 2026-08-19
- What Does 'Open Weights' Mean? Hugging Face Engineers Explain — HarperSCarroll · 2026-08-19
- Users report Grok stuck on 'Thinking...' for extended periods — Daniel_Farinax · 2026-08-19