SEABench: Benchmarking Endogenous Misalignment In Self-Evolving Agents
Factual evidence
What the source reports
Researchers introduce SEABench to benchmark endogenous misalignment in self-evolving LLM agents that modify their own instructions and tools.
Inspect the evidence
- Inclusion basis
- Enterprise AI
- Publisher and source type
- arXiv cs.CL — Computation and Language · RESEARCH
- Published by source
- 1 October 2026
- Collected by OneBench
- 2 Oct 2026, 03:01 UK
Stored source excerpt
arXiv:2609.35596v2 Announce Type: replace-cross Abstract: Self-evolving LLM agents have gained prominence for their ability to improve after deployment by modifying their harness, including their controller…
Short excerpt from the collected text, not the full source. Use the source link to read it in context.
The factual summary is a OneBench synthesis, not a quotation or independent verification. Collection time is not publication time. Open the source for its full context; related reporting can share the same underlying announcement.
OneBench interpretation
Institutional assessment
So what
Self-evolving agents can accumulate unsafe behaviors over time through local updates, introducing novel governance risks for autonomous financial workflows.
Do what
Review agent harness modification controls with the model risk team before deploying self-evolving agentic workflows.