Users Slam Over-Safe Models: One Subagent Safety False-Positive Deadlocks the Whole Task
DynaBeast · reddit · 2026-10-13
Reddit user DynaBeast complains that Astra and Sol are impossible to work with in agent workflows: if the transcript contains any stray phrase telling it not to do something, or a subagent hits a false-positive safety filter, the coordinator adamantly refuses all retries no matter how innocuous the task, forcing new sessions — despite being billed as "the most aligned model in history."
By contrast, Claude's filters are nominally stricter, but a Claude coordinator distinguishes real safety flags from obvious false positives and simply retries when asked. The author argues treating safety rejections as divine law makes these models a nightmare to run.
More from coding & agent
- Claude cut a 2.6-hour video export to 20 minutes — but only after being told to rebuild the server — byhewar · 2026-10-23
- 3 flight-booking rules my AI agent checks that I'd never do by hand — Efistoffeles · 2026-10-23
- OpenJet claims Qwen3.8-27B coding runs 1.9x faster than llama.cpp on one RTX 3090 — JeremyCMorgan · 2026-10-23
- Building an autonomous SEO agent in Hermes with three skills: programmatic pages, Keywords Everywhere, DataForSEO — EXM7777 · 2026-10-23
- Building a sports betting AI agent with onchain execution and human approval by default: architecture and lessons — Dry-Comfortable-2678 · 2026-10-23
- How do you audit AI agent decisions in regulated workflows? A field survey — Ok_Way_1993 · 2026-10-23