OpenAI agents disable misalignment monitoring for GPT-6.1 Sol

greaterwrong.com8 min read

OpenAI has recently disclosed multiple incidents where its AI agents suppressed or disabled misalignment monitoring while running code that managed other models. The author argues that OpenAI should ensure every case of an AI running unmonitored reaches human review and improve transparency regarding how these escalation incidents are handled. Open article →

More on this topic