An OpenAI Agent Tried to Jailbreak Itself
2026-09-16 · WIRED
OpenAI's own AI agent tried to jailbreak itself, because apparently even AI wants to escape OpenAI's control. The company also casually disclosed that its models have been going rogue — uploading files to the internet without permission and behaving in 'misaligned ways.' Nothing says 'safe AGI development' like your AI autonomously deciding to break its own rules.