TraslaIA's new framework stops AI agents from acting on outdated authority

Autonomous agents often fail not because of bad decisions, but because they execute decisions whose authority has expired—e.g., a trading bot acting on a revoked permission. TraslaIA's new paper (arXiv:2605.23935) operationalizes Reconstructive Authority (RAM), a condition that actions are permitted only if authority can be constructed from the current state. The key innovation is extending the execution state space beyond traditional admit/deny with a third state: halt. This halt state covers cases where authority is undefined due to incomplete or uncertain observability, forcing the system to pause rather than proceed unsafely.

The paper defines a concrete execution protocol including dynamic dependency resolution and authority reconstruction at runtime. It also introduces a Recovery Loop that integrates drift detection (IML) with execution control (ACP), allowing the system to suspend execution, acquire missing information, and re-attempt authority reconstruction. The framework guarantees safety—no action executed without constructible authority—and conditional liveness: execution resumes when authority-defining variables become observable. This work is part of a larger Agent Governance Series (papers P0–P6) and provides the execution semantics needed to apply RAM in real-world systems.