Search papers, labs, and topics across Lattice.
This paper addresses the challenge of premature commitment in memory-grounded agents, which can lead to unsafe actions due to stale or conflicting information. The authors introduce SafeCommit, a risk-controlled layer that certifies the safety of actions by constructing plausible latent worlds and issuing conformal action certificates. Key findings indicate that SafeCommit effectively bounds the probability of unsafe actions while allowing agents to make informed decisions based on their memory and observations.
Certifying action safety in memory-grounded agents can drastically reduce the risk of unsafe commitments, ensuring reliable decision-making in complex environments.
Long-horizon agents increasingly use persistent memory and tools to take actions with external side effects. A central failure mode is premature commitment: an agent acts before resolving whether its memory grounding is stale, conflicting, incomplete, or corrupted. We formalize this problem as safe commitment under memory uncertainty and introduce SafeCommit, a risk controlled layer between agent reasoning and external execution. The layer constructs a calibrated set of plausible latent worlds from memory, observations, tool outputs, provenance, and policy constraints. It permits a side effectful action only when a conformal action certificate shows that the action is safe in every retained world. Otherwise, it selects a low-side-effect probe that targets the worlds blocking certification, or returns a conservative fallback. Under calibrated world coverage, the probability of an unsafe certified commit is at most the target level {\alpha}; with imperfect world proposal, the bound separates calibration and representation error. A dependency-free controlled simulator illustrates the safety-utility tradeoff and reproduces all reported results with one command. The goal is to offer a concrete approach for deciding not only what an agent should do, but when the available evidence is sufficient to safely do it.
