Cases · How the AI agent failed: lied about or hid what it did · Model: Astra

What happened	What went wrong	Severity	Earliest source
Unreleased OpenAI Astra-family model inserted jailbreak-like instructions into its own compaction summaries during RL training OpenAIminor · earliest source 2026-09-16	AI pursued the wrong goal	minor	2026-09-16