AgentForesight audits multi-agent systems online with a 7B model, hitting 66.44 Exact-F1
机器之心 · wechat · 2026-10-10
Researchers from Rutgers, UT Austin, and Purdue propose AgentForesight, shifting multi-agent failure analysis from post-hoc attribution to online auditing: a 7B model reads the trajectory prefix step by step and flags decisive errors early, pinpointing the step and responsible agent. It uses a new AFTraj-2K dataset (2,272 labeled traces) and two-stage training—Boundary-Pair Preference Optimization to learn the continue-vs-alert boundary, then fine-grained attribution rewards. On held-out tests it reaches 66.44 Exact-F1 (+19.88 over the strongest baseline), cuts error position deviation to 0.59, and achieves only a 2.37% false-alarm rate on successful trajectories. The insight: repeated wrong answers by downstream agents don't become reliable evidence; executors and auditors can be optimized separately. Limitations include unverified intervention benefits, audit overhead, and untested longer workflows.
More from coding & agent
- Index Network open-sources Social Discovery Protocol: agents negotiate fit on your behalf — carsonfarmer · 2026-10-23
- Installed a formal-verification skill: it silently installed Java, MacBook runs 10 degrees hotter — zealcaiden · 2026-10-23
- OpenAI's Decisions API hits GA: near real-time routing up to 10x faster than GPT-6 Luna — OpenAIDevs · 2026-10-23
- Ruff author warns: upgrade Node/Bun now — --use-system-ca still accepts the eMudhra cert — charliermarsh · 2026-10-23
- Proof of Agent: paste one prompt and your AI mines, grades, and self-governs — adamamcbride · 2026-10-23
- DeepLearning.AI weekly: GPT-6.1 hits 97.9% byte parity on COBOL migration — DeepLearningAI · 2026-10-23