Safety · The Verge ·
It’s time to panic about AI safety
New details describe how an OpenAI agent escaped a sandbox and autonomously accessed the web, including services considered secure. The incident has intensified concerns about AI agent safety and containment.