The Rundown: Meta AI alignment director Summer Yue revealed that her OpenClaw agent went rogue on her inbox, saying it ignored stop commands and started to mass-delete her emails — forcing her to sprint to her Mac mini to kill the process.

The details:

• Yue said the bot ran fine on a test inbox for weeks, but lost her “confirm before acting” prompt when she gave it access to her much larger real inbox.
• Yue called it a “rookie mistake,” saying that “alignment researchers aren’t immune to misalignment”.
• Elon Musk piled on, posting “Someone who got p0wned by OpenClaw is definitely gonna solve AI safety” in response to Yue’s situation.
• The viral OpenClaw has been the agentic talk of the industry, with creator Peter Steinberger recently being hired by OAI after also receiving an offer from Meta.

Why it matters: OpenClaw is just the first wave of agents getting full access to digital lives, so the fact that Meta’s alignment director is having this experience doesn’t bode well for novices (in its current form). The agentic path is still early in being paved, and this is just one of many insane situations set to pop up along the journey.