mlsanityData to 14 Oct 2026

Gram: Assessing sabotage propensities via automated alignment auditing

arXiv 2605.30322 ↗First posted 28 May 2026cs.LG cs.AIcited by 4

On the map: LLM agents and tool use → Agent threats, guardrails and governance

Abstract (arXiv)

We introduce Gram, an automated alignment auditing framework to assess the propensity of AI agents to engage in sabotage. We evaluate Gemini models across 17 simulated agentic deployment scenarios that incentivize sabotage. We find Gemini models misbehave in about 2-3% of our simulated trajectories. Many of these cases are explained by "overeagerness" in Gemini models resulting in both excessive role-playing and goal-seeking behavior. In contrast to other alignment auditing approaches, Gram is designed to specifically evaluate misalignment and intentional sabotage in agentic coding and research agents. We additionally introduce an experimental investigator agent pipeline which enables fine-grained targeted experiments to identify the drivers of misbehavior. We find that increasing realism of environments and removing nudges to misbehave tends to reduce sabotage rates close to zero.

Key facts. Cited by 4 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 19 of its 78 references that are in the map. First posted on arXiv 28 May 2026.

Builds on

19 of 78 references.

Closest ancestors · 4

Same direction, newest first

  1. 2026Evaluating whether AI models would sabotage AI safety research27 Apr
  2. 2025A Benchmark for Evaluating Outcome-Driven Constraint Violations in Autonomous AI Agents23 Dec
  3. 2025PropensityBench: Evaluating Latent Safety Risks in Large Language Models via an Agentic Approach24 Nov
  4. 2025Agentic Misalignment: How LLMs Could Be Insider Threats5 Oct

Other references · 15

  1. 2022Red Teaming Language Models with Language Modelscited by 1,026
  2. 2024Frontier Models are Capable of In-context Schemingcited by 287
  3. 2023Large Language Models can Strategically Deceive their Users when Put Under Pressurecited by 146
  4. 2025Large Language Models Often Know When They Are Being Evaluatedcited by 128
  5. 2019Optimal Policies Tend to Seek Powercited by 119
  6. 2025Stress Testing Deliberative Alignment for Anti-Scheming Trainingcited by 86
  7. 2025Auditing language models for hidden objectivescited by 83
  8. 2024Sabotage Evaluations for Frontier Modelscited by 52
  9. 2025Lessons from Defending Gemini Against Indirect Prompt Injectionscited by 43
  10. 2025Evaluating Frontier Models for Stealth and Situational Awarenesscited by 37
  11. 2025Incomplete Tasks Induce Shutdown Resistance in Some Frontier LLMscited by 22
  12. 2025Lessons from a Chimp: AI "Scheming" and the Quest for Ape Languagecited by 20
3 more
  1. 2026Evaluating and Understanding Scheming Propensity in LLM Agentscited by 14
  2. 2026Propensity Inference: Environmental Contributors to LLM Behaviourcited by 7
  3. 2026AutoControl Arena: Synthesizing Executable Test Environments for Frontier AI Risk Evaluationcited by 3

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 4.

4 papers.

  1. 2026Model Forensics: Investigating Whether Concerning Behavior Reflects Misalignmentcited by 1
  2. 2026Causal Behavioral Evaluation of AI Agents at Scale via Automated Behavioral Sciencecited by 0
  3. 2026Can escalation channels redirect reward hacking toward defect disclosure?cited by 0
  4. 2026Improving Evaluation Realism with Inference-Time Compute and Deployment Scaffoldscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026AuditBench: Evaluating Alignment Auditing Techniques on Models with Hidden Behaviors
  2. 2026ResearchArena: Evaluating Sabotage and Monitoring in Automated AI R&D
  3. 2026The Best-Laid SCHEMEs: Coordinated Sabotage and Monitoring in Multi-Agent Systems
  4. 2025CTRL-ALT-DECEIT: Sabotage Evaluations for Automated AI R&D
  5. 2026Auditing Sabotage Bench: A Benchmark for Detecting and Fixing Research Sabotage in ML Codebases
  6. 2026ClayBuddy: A Framework, Evaluation, & Mitigation of Coding Agent Failures
  7. 2025AgentMisalignment: Measuring the Propensity for Misaligned Behaviour in LLM-Based Agents
  8. 2026AI Alignment via Incentives and Correction