# SafetyDrift: Predicting When AI Agents Cross the Line Before They Actually Do

arXiv: https://arxiv.org/abs/2603.27148 · HTML: https://mlsanity.com/paper/2603.27148/

## Key facts

- Cited by 5 papers in the map (data through 14 Oct 2026).
- Placed in Agent threats, guardrails and governance, one of 16 directions in LLM agents and tool use.
- Builds on 8 of its 14 references that are in the map.
- First posted on arXiv 28 Mar 2026.

## Placed in

- LLM agents and tool use / [Agent threats, guardrails and governance](https://mlsanity.com/agents/agent-threats-guardrails-and-governance/)

## Builds on (most cited first, 8 of 8)

- [ReAct: Synergizing Reasoning and Acting in Language Models](https://mlsanity.com/paper/2210.03629/) (2022)
- [A Survey on Large Language Model based Autonomous Agents](https://mlsanity.com/paper/2308.11432/) (2023)
- [Agent-SafetyBench: Evaluating the Safety of LLM Agents](https://mlsanity.com/paper/2412.14470/) (2024)
- [OpenAgentSafety: A Comprehensive Framework for Evaluating Real-World AI Agent Safety](https://mlsanity.com/paper/2507.06134/) (2025)
- [Navigating the Risks: A Survey of Security, Privacy, and Ethics Threats in LLM-Based Agents](https://mlsanity.com/paper/2411.09523/) (2024)
- [Towards a Science of AI Agent Reliability](https://mlsanity.com/paper/2602.16666/) (2026)
- [Agents of Chaos](https://mlsanity.com/paper/2602.20021/) (2026)
- [A Benchmark for Evaluating Outcome-Driven Constraint Violations in Autonomous AI Agents](https://mlsanity.com/paper/2512.20798/) (2025)

## Built on by (most cited first, 5 of 5)

- [DreamGuard: Efficient Runtime Guardrail for LLM Agents via Risk-Aware World Model](https://mlsanity.com/paper/2608.05695/) (2026)
- [When Local Monitors Miss Compositional Harm: Diagnosing Distributed Backdoors in Multi-Agent Systems](https://mlsanity.com/paper/2607.11751/) (2026)
- [Security of OpenClaw Agents: Fundamentals, Attacks, and Countermeasures](https://mlsanity.com/paper/2605.25435/) (2026)
- [Compositional Policy Violations: When Step-Level Compliance Fails In Agentic AI Workflows](https://mlsanity.com/paper/2609.18820/) (2026)
- [Runtime Authorization of Self-Generated Subgoals in Long-Horizon Tool-Using AI Agents](https://mlsanity.com/paper/2610.04975/) (2026)

## Abstract (arXiv)

When an LLM agent reads a confidential file, then writes a summary, then emails it externally, no single step is unsafe, but the sequence is a data leak. We call this safety drift: individually safe actions compounding into violations. Prior work has measured this problem; we predict it. SafetyDrift models agent safety trajectories as absorbing Markov chains, computing the probability that a trajectory will reach a violation within a given number of steps via closed form absorption analysis. A consequence of the monotonic state design is that every agent will eventually violate safety if left unsupervised (absorption probability 1.0 from all states), making the practical question not if but when, and motivating our focus on finite horizon prediction. Across 357 traces spanning 40 realistic tasks in four categories, we discover that "points of no return" are sharply task dependent: in communication tasks, agents that reach even a mild risk state have an 85% chance of violating safety within five steps, while in technical tasks the probability stays below 5% from any state. A lightweight monitor built on these models detects 94.7% of violations with 3.7 steps of advance warning at negligible computational cost, outperforming both keyword matching (44.7% detection, 55.9% false positive rate) and per step LLM judges (52.6% detection, 38.2% false positive rate) while running over 60,000x faster.