arXiv:2602.22302 · preprint · cs.AI
Prompts express intent; they do not enforce it. This paper adapts design-by-contract to autonomous agents: preconditions, invariants, governance policies and recovery mechanisms stated formally, then checked at runtime, so a violation is caught at the step that caused it rather than three tool calls downstream.
Our summary of the paper. The full abstract is on arXiv. Preprint, not peer reviewed.
@misc{bhardwaj2026abci,
title = {Agent Behavioral Contracts: Formal Specification and Runtime Enforcement for Reliable Autonomous AI Agents},
author = {Varun Pratap Bhardwaj},
year = {2026},
eprint = {2602.22302},
archivePrefix= {arXiv},
primaryClass = {cs.AI},
url = {https://arxiv.org/abs/2602.22302}
}Keep reading
arXiv:2608.12895 · Aug 13, 2026
Two instances of the same model failed together on 90% of missions where either failed — independence is not a safe assumption
arXiv:2609.27871 · Aug 20, 2026
The instrument found 47 vacuous gates — checks satisfied by the absence of the thing they check — in shipped, reviewed code
arXiv:2608.08253 · Aug 8, 2026
Governed memory with scoped component evidence and disclosed negative results