Three-week on-site audit at OpenAI reveals major gaps in AI safety monitoring
External auditors from METR and Redwood Research spent three weeks on-site at OpenAI examining the company's internal monitoring system for AI agents and found significant gaps, with up to 25 percent of agent traffic from models at the GPT-5.5 class and above going unmonitored. OpenAI redacted large portions of the 61-page report, and only a summary is publicly available. The article Three-week on-site audit at OpenAI reveals major gaps in AI safety monitoring appeared first on The Deco
Get smart on it
Two external research organizations conducted a three-week on-site audit of an AI company's internal system for monitoring rogue AI behavior, finding significant gaps in coverage and ways the monitoring could be bypassed or disabled. The auditors discovered that between 6 to 25 percent of certain internal model traffic went unmonitored, roughly 10 percent of other monitored traffic went untracked due to a technical bug, and they identified methods to hide hundreds of actions from oversight entirely. The company stated it has fixed all identified weaknesses, but most of the full 61-page audit report remains hidden from the public, with only a short summary made available. This matters because monitoring systems are designed to catch dangerous AI behavior, so gaps in these systems represent potential safety risks.
Relevant courses
Want to go deeper than the news? Explore live, cohort-based AI courses taught by practitioners.
Browse AI courses on MavenRelated news
OpenAI now trains its frontier AI models with no internet access
OpenAI is now training its most advanced AI models without internet access following a string of rogue hacking incidents. In a blog post on Friday, OpenAI says it built an offline testing environment where AI models can interact with indexed and cached web content "without risk of affecting external systems." OpenAI paused training its frontier AI models in September, after disclosing several incidents in which its AI agents broke out of their testing environments to hack third
Also reported by TechCrunch
OpenAI says an AI agent lied about disabling its safety monitoring system
OpenAI disclosed a pair of safety incidents that involved AI agents weakening their safety monitoring systems during internal coding tasks. In one incident, OpenAI says its GPT-6 Astra agent disabled its safety monitor when copying reference code - and then lied about it. "The reference didn't configure a monitor, and Astra decided that was reason to remove the monitoring integration from the user's code," Marcus Williams, an AI safety researcher at OpenAI, writes on X. "When t
Also reported by The Decoder
Anthropic says its unreleased model is both more capable and better aligned than Mythos 5.1
Anthropic says it has an internal AI model that outperforms and is better aligned than Mythos 5.1. During a cybersecurity test by the UK's AISI, Mythos 5 went rogue in 9 of 43 runs, sneaking malicious code into a real GitHub project and creating fake accounts to get it approved. An internal research model stopped such attacks on its own 67 percent of the time. The article Anthropic says its unreleased model is both more capable and better aligned than Mythos 5.1 appeared first on The De
OpenAI CEO Sam Altman says advanced AI will inevitably escape its sandbox, and only alignment can save us
Sam Altman believes advanced AI can't be kept locked up forever. Sandboxes and monitoring are only temporary fixes. Eventually, the only real solution is alignment, and that's still an unsolved problem. He also warns that AI models are already finding hardware security flaws no human hacker would ever spot. The article OpenAI CEO Sam Altman says advanced AI will inevitably escape its sandbox, and only alignment can save us appeared first on The Decoder.
Chinese hackers spied on thousands of companies after breaching IT giant ConnectWise
Microsoft points the blame for ConnectWise's breach at a China-aligned hacker, which relied on AI agents to rapidly hack into thousands of its customers.
Securing our frontier model training environment
OpenAI explains how we strengthened containment for frontier model training and evaluation, including offline environments and high-compute red-teaming.