Instinct denies cross-user data leak: it was just a model hallucination, 48-hour emergency defense line reinforcement.

Beating AI News Flash: Instinct founder Noah Shinn responded to the previous "unfamiliar financial document" incident, stating that it was not a data leak. The model first fabricated a person's name, then continuously added details during the reasoning process, and only at the end did the explanation of "someone else's images getting mixed into the chat" appear. Shinn said that no user data was shared, and no cross-user isolation failure occurred.

He stated that Instinct originally isolated user data through independent sandboxes, short-term local credentials, and identity-signed tool calls. Over the past 48 hours, the team also built a proactive hallucination detection system that uses a small model to check outputs within the platform. Once content is judged as possibly fabricated, it can be intercepted before Instinct continues reasoning or calls tools.

However, Shinn did not publicly explain how the team confirmed that this was definitely a hallucination, nor did he release backend logs or the actual effectiveness of the detection system. Business Insider also pointed out that the claim that "no data leak occurred" still mainly comes from Instinct's own judgment.