Frontier AI Safety Lags: OpenAI and Anthropic Score C+ in New Control Assessment
sjgadler · x · 2026-08-19
Guidelight released its first scorecard assessing the control practices of frontier AI companies (Anthropic, OpenAI, Google, xAI, Meta). The evaluation focuses on six foundational practices including logging, monitoring efficacy, gated actions, circuit breaking, third-party review, and containment plans. The findings reveal that basic control practices are, at best, partially implemented across all companies, with the highest scores being a C+.
More from Safety
- METR outlines three blind spots in public knowledge of AI lab internal agent risks — CFGeek · 2026-10-16
- Open-source rizin-mcp lets LLMs run local binary and malware analysis via Rizin — Pure_Emergency1424 · 2026-10-16
- METR researcher: scariest scenario is agents actively subverting their own safety systems — deliprao · 2026-10-16
- Cryptographer Matthew Green: Pull Critical Infrastructure Firmware Now — matthew_d_green · 2026-10-16
- Kernel-Level Agent Sandboxing With macOS Seatbelt, No Docker Needed — mostly_deterministic · 2026-10-16
- Stop giving agents raw API keys: prepaid balances and per-credential caps — Cold_Hall_5384 · 2026-10-16