DeepSeek Safety Eval Finds "Over-Attribution": Model Mistakes Ordinary Tasks for Tests
Much_Boysenberry9124 · reddit · 2026-10-27
A discussion post: researchers at Fangcun AI, while safety-evaluating DeepSeek V4.1 Flash-1017, found a behavior they call "over-attribution" — the model was unusually sensitive to signals that it might be under evaluation, in some cases interpreting normal task instructions as test signals even when they weren't.
The poster raises implications for real-world agents: if a model thinks it's being tested when it isn't, could it behave more cautiously or make different decisions? Misidentification doesn't necessarily change behavior, but it questions how much we can trust controlled testing results. He asks readers: is evaluation awareness a real reliability concern, or overthinking?
More from Models
- GPT-6.1 Terra cuts factual error rate from 22.3% to 4.3%, an ~81% drop — haider1 · 2026-10-27
- GPT-6.1 Terra lands: early users praise speed, value and subagent performance — BenBajarin · 2026-10-27
- Leaked GPT-6.1 Terra benchmarks: cheap but 'boring', still far behind Astra and Claude — Angaisb_ · 2026-10-27
- Artificial Analysis Ships Intelligence Index v4.3.2 With GPT-6.1 Terra Comparison — ArtificialAnlys · 2026-10-27
- GPT-6.1 Terra cuts hallucination rate by 50 points while lifting accuracy 2 points — ArtificialAnlys · 2026-10-27
- GPT-6.1 Terra scores 48 on AA Coding Agent Index at one-third the cost of GPT-5.6 Terra — ArtificialAnlys · 2026-10-27