GPT — Field Evidence
paperUnverified
GPT · GPT · incident
Field note
Institutional Red-Teaming: Deployment Rules, Not Just Models, Causally Shape Multi-Agent AI Safety — We introduce institutional red-teaming, an evaluation methodology for testing deployment rules in multi-agent AI: hold the agents, objectives, and task state fixed, vary only one
collected 2026-07-09original 2026-07-08
Does this shift the US–China race?
Be the first to call it
Impact on the index
Safety concern · minor
US -15
Directional contribution — before recency decay and per-type diminishing returns. How it works →
Related Front
US vs ChinaLikely
U.S. FrontiervsChina Open-Weight
Likely — capability gap narrowing on common tasks
Letters from the Front
Be the first to file a report.