TMLR paper: enforced thinking doesn't reliably make LLM agents disclose key info to users
SharonYixuanLi · x · 2026-10-27
A University of Wisconsin–Madison and William & Mary team (with Jindong Wang and Sharon Li) published a TMLR paper, "Thinking Is Not Telling: Information Disclosure in User-Service LLM Agents."
- Core problem: service agents see policies, tool results, and environment states that users can't; users need this info before consequential decisions, but private reasoning doesn't guarantee timely disclosure.
- Methods: a five-category information-disclosure taxonomy, evidence-grounded probing, and turn-level counterfactual interventions to study what agents communicate and how missing information affects their actions.
- Findings: across 5 models and 3 service scenarios, enforced thinking does not consistently improve task success, and more thinking doesn't reliably yield richer disclosure. The paper's example shows a non-thinking agent proactively flagging the key policy that only whole orders can be cancelled.
Paper and arXiv links are live; code is coming soon.
More from coding & agent
- Reticle builds a verification layer that lets AI coding agents check their own generated code — Scobleizer · 2026-10-27
- Trust your agent, but read its logs: Claude managed agents tips — brada · 2026-10-27
- Dev training a guard model to block prompt injection, tool and memory poisoning — gnukeith · 2026-10-27
- Dev trains a ~4GB-VRAM sentinel model to guard against prompt injection and poisoned memory — gnukeith · 2026-10-27
- Superset rebrands as Paradev, raises $11.5M seed led by USV — massimosgrelli · 2026-10-27
- Ex-Big Tech managers may thrive in the AI era: manage agents like you managed teams — lxfater · 2026-10-27