User alleges Claude silently compresses context early and blocks git commits mid-session
maier_ak · x · 2026-09-15
@maierak alleges more opaque Claude UI behavior: a green "action-allowed" toggle can allegedly flip to red mid-session, blocking simple git commits without warning, and chat history is claimed to be silently compressed before the advertised context limit, leaving users unaware of lost context. The post links to a critique titled "Anthropic's Harness Audits the Fire Extinguisher, Ignores the Gas Leak".
Unverified single-user claims, no official confirmation yet.
More from Safety
- UK's 'AI sovereignty' push undermined by its own energy and copyright policy — NinaDSchick · 2026-11-02
- Auditing my MCP credential broker: the worst hole was the config file — Relevant_Mousse_9117 · 2026-11-02
- Cambricon co-founder urges shared US-China AI ethics standards for leading models — pstAsiatech · 2026-11-02
- Vitalik's "post-AI cryptography": pseudonyms are dead, per-action anonymity is in — devanshmehta · 2026-11-02
- 6 prompts to see what ChatGPT quietly knows about you — and wipe it — Aiden_Tech_Ai · 2026-11-02
- European supermarket freezers hit by second suspected cyber-attack — fergie · 2026-11-02