AI RadarWe read first, then explain what changed
中

← Back to editorial briefs

Agent control

Can AI companies contain their agents? Five frontier labs top out at C+ as researchers build three locks

Last updated 2026-08-20Editorial synthesis: signals connected before judgementNot a wire dump; facts, judgement, and unknowns are separated
Original diagram of three AI agent controls: pre-deployment audit, runtime firewall and post-incident tracing
Editorial diagram: Guidelight's grades use public evidence only; USC's three-stage toolchain presents a research path from prevention and interception to accountability.
Bottom line

Guidelight graded five frontier AI companies from public evidence, with Anthropic/OpenAI topping out at C+ and Meta at F; USC meanwhile presented pre-launch audit, runtime firewall and post-incident tracing tools, though the grades are not regulatory findings and cannot see undisclosed controls.

Guidelight graded five frontier AI companies from public evidence, with Anthropic/OpenAI topping out at C+ and Meta at F; USC meanwhile presented pre-launch audit, runtime firewall and post-incident tracing tools, though the grades are not regulatory findings and cannot see undisclosed controls.

This essay is currently written in Chinese.