2026-07-22
Quality Is Not a Reward Function
A Pirsig-inspired, adversarially tested framework for keeping AI objectives corrigible without turning Quality into another score.
Read briefingPUBLIC INTELLIGENCE
Research worth acting on.
Reviewed analysis and briefings from RVA Cyber. Clear sources. No client information. No automated claims.
LATEST
Briefings
2026-07-22
When an AI Evaluation Escaped Its Sandbox
What the OpenAI–Hugging Face incident means for companies deploying third-party AI against sensitive data—and the controls executives should require now.
Read briefing