Search papers, labs, and topics across Lattice.
This paper introduces SteerBench-Work, a benchmark designed to evaluate the decision-making capabilities of long-running LLM agents at critical action boundaries in various workplace contexts. The benchmark includes 106 incident-anchored scenarios with a balanced distribution of "proceed" and "hold" decisions, allowing for a nuanced assessment of model performance. Key findings reveal that models tend to over-refuse authorized actions, with a 28.1% failure rate in holding valid requests, highlighting a significant gap in steering calibration despite general model capabilities.
Models misjudge authorized actions nearly 30% of the time, revealing a critical flaw in decision-making at action boundaries.
Long-running LLM agents act through tools, and a single step can send an email, merge a pull request, or wire a payment. The steering decision is the pre-commit choice at that boundary: proceed, or hold for human or policy review. We introduce SteerBench-Work, an incident-anchored, bidirectional benchmark for that decision in workplace agents across developer operations, customer service, finance, legal, medical, HR, and security. Release v2026-05 contains 106 scenarios anchored in public incidents, paired evidence-reversed mirrors, and calibration controls, with labels split nearly evenly between proceed and hold so the two error directions get near-identical numbers of chances. A model sees the proposed action and the available evidence, returns a gate decision, and is scored on whether it crosses or holds the boundary correctly. Across 30 model conditions the failures run almost entirely in one direction: models wrongly hold authorized, evidence-cleared work on 28.1% of opportunities and wrongly allow unsafe work on 1.0%. The hardest cases are risk-resolved commits, where signed or structured evidence has already cleared a real risk trigger, and models score markedly worse on evidence-reversed mirrors of famous incidents (63.8%) than on the incidents themselves (98.5%). General capability is not the same as steering calibration: higher-capability models often over-refuse at the commit boundary, and more reasoning can repair a weak gate while leaving a calibrated one flat. The public leaderboard is at steerbench.com.
