evidence blindness
Having access to relevant evidence does not prevent an LLM agent from making incorrect or unsafe decisions.
- Papers
- 10
- Released code
- 2
- First seen
- July 2026
- Latest
- Sept 2026
9 papers in the last two months, against 1 in the two before.
Who is working on it
The papers
Most central to this idea first, not most recent.
- Research Labcs.CR
Execution Logs Can Mislead Visual Judges in Video-Generation Agents
RIKEN · Sept 2026
- Top Universitycs.CR
Agent approvals can omit effects triggered downstream by tools
Institute of Information Engineering, Chinese Academy of Sciences, University of Chinese Academy of Sciences · Sept 2026
- Independentcs.AI
Polished evidence makes LLM agents act on unknowable questions
Aug 2026
- Top Universitycs.CR
Agent safety requires persistent state across autonomous loop iterations
University of Chinese Academy of Sciences, Nanyang Technological University · Aug 2026
- Independentcs.AI
Persistent corpus maps help agents find evidence within tight budgets
Aug 2026
- Independentcs.SE
Multi-image evidence can help models repair software, but unreliably
Sept 2026
- Chinese Techcs.AI
Restructuring agent environments improves performance on noisy, evolving tasks
Shanghai Jiao Tong University, Theseus Labs · Sept 2026
- Top Universitycs.AI
Financial agents can cite rules while still attempting prohibited trades
HKUST, HKBU · Aug 2026
- Chinese Techcs.AIcode
C3M preserves multimodal evidence across sessions within fixed memory budgets
Huzhou Normal University, Alibaba Group · Sept 2026
- Chinese Techcs.AIcode
Automated safety testing reveals 93.9% attack success rate across four agent frameworks
AntGroup, Zhejiang University · July 2026
Concepts are extracted from each paper and reused across the corpus, so this page grows on its own as the desk reads.