R0044/2026-04-01/Q003/SRC01
Ibrahim et al. (2025) — Measuring and mitigating overreliance is necessary for building human-compatible AI
Source
| Field |
Value |
| Title |
Measuring and mitigating overreliance is necessary for building human-compatible AI |
| Publisher |
arXiv (multi-institutional) |
| Author(s) |
Lujain Ibrahim (Oxford), Katherine M. Collins (Cambridge), Sunnie S. Y. Kim (Princeton), Anka Reuel (Stanford/Harvard), Max Lamparth (Stanford), Kevin Feng (UW), Lama Ahmad (OpenAI), et al. |
| Date |
2025 |
| URL |
https://arxiv.org/html/2509.08010v1 |
| Type |
Research paper (preprint) |
Summary
| Dimension |
Rating |
| Reliability |
Medium-High |
| Relevance |
High |
| Bias: Missing data |
Low risk |
| Bias: Measurement |
Low risk |
| Bias: Selective reporting |
Low risk |
| Bias: Randomization |
N/A — not an RCT |
| Bias: Protocol deviation |
N/A — not an RCT |
| Bias: COI/Funding |
Some concerns |
Rationale
| Dimension |
Rationale |
| Reliability |
Multi-institutional team from Oxford, Cambridge, Princeton, Stanford/Harvard, and OpenAI. Preprint (not yet peer-reviewed) but comprehensive review with original framework. |
| Relevance |
The single most relevant source for Q003 — explicitly discusses automation bias, sycophancy, trust calibration, and cognitive offloading in a unified framework. This is the closest thing to a vocabulary bridge found. |
| Bias flags |
Some COI concern: Lama Ahmad is from OpenAI, which has a commercial interest in sycophancy mitigation framing. However, the paper's multi-institutional authorship mitigates this. |
| Evidence ID |
Summary |
| SRC01-E01 |
Unified framework connecting cognitive science, HCI, and AI safety concepts around overreliance |