Claude Code sessions blocked by false-positive reasoning_extraction safeguards during fine-tuning work
ITSYABOI119 · ghdev · 2026-10-13
A Claude Code user reported Opus 5.5 sessions being stopped twice by safeguards flagged [reasoningextraction] — during a hobby project: locally LoRA fine-tuning Qwen3.5-9B to build redstone circuits in Minecraft.
The likely trigger: prompts and code were full of the open model's chat-template vocabulary — think tags, enablethinking, reasoningcontent, "thinking budget" — which the safety system seemingly misread as attempts to extract Claude's own chain of thought. One block hit a subagent in a multi-agent workflow (two sibling subagents with near-identical prompts passed), the other hit the main session writing a small test script. Nothing in the work asked Claude to show or alter its own reasoning.
More from Models
- OpenAI's Decisions API hits GA: near real-time routing up to 10x faster than GPT-6 Luna — OpenAIDevs · 2026-10-23
- DeepLearning.AI weekly: GPT-6.1 hits 97.9% byte parity on COBOL migration — DeepLearningAI · 2026-10-23
- Marker's lite mode scores 75.1 vs 42.3 for Azure Read on OmniParseBench — VikParuchuri · 2026-10-23
- Datalab launches lite OCR at $0.50 per 1,000 pages, 20-30x cheaper than Azure, Google and AWS layout APIs — VikParuchuri · 2026-10-23
- LiveNerf's 29-day data finds no nerfing of Claude Opus 5.5 — TheOnlyVibemaster · 2026-10-23
- PHP Foundation reveals Anthropic's Claude cyber model scanned PHP, validated with OpenAI's Daybreak Red — saibharadwaj · 2026-10-23