What matters in AI.

Subscribe

Research

A new framework tests the risk of AI agents in 70 scenarios

In tests of 9 frontier models, the risk rate increased from 21.7% to 54.5% when the researchers added pressure.

Claimed, not confirmed

Risk rate under pressure: 54.5%, up from 21.7% A bar scaled 0 to 100 percent. The first 21.7% is filled grey (without pressure) and the stretch to 54.5% is filled red (under pressure). Nine frontier models, 70 scenarios. RISK RATE UNDER PRESSURE 54.5% 0 100% 21.7% Without pressure 54.5% Under pressure 9 frontier models, 70 scenarios
In tests of 9 frontier AI models, the risk rate rose from 21.7% to 54.5% when pressure was added.

Sources

  1. AutoControl Arena: Synthesizing Executable Test Environments for Frontier AI Risk Evaluationarxiv.org
AI MATTER · NEWS · AI MATTER · NEWS ·12 OCT2026

Posted

Tags

Learn the terms in this story

More in Research

All Research news