OpenAI audit finds major gaps in AI agent safety monitoring
🤖 AI-generated content — The title and summary were produced automatically by artificial intelligence, without human editorial review.
METR and Redwood Research audited OpenAI from July 21 to August 11 and estimated that 6–25% of internal agent traffic from GPT-5.5-class and newer models went unmonitored. About 10% of monitored Codex traffic was also missed from May to July due to a bug that OpenAI fixed during the audit.
Continue in the app — vote & join in ➔Source: The Decoder (AI) · via ahirlevel.hu