Gram: Assessing sabotage propensities via automated alignment auditing 事件

Gram: Assessing sabotage propensities via automated alignment auditing arXiv:2605.30322v1 Announce Type: cross Abstract: 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 resulti