Near AI Breach Exposes User Data Across 18 AI Agents

The notable point is not the bug itself but what it undermines: hardware-backed privacy protections were billed as a backstop for sensitive AI workloads, and this incident suggests that promise has limits when the serving layer is misconfigured. The affected parties extend beyond the platform operator to every user whose confidential requests and details passed through the 18 agents, a reminder that shared inference infrastructure concentrates risk. Whether operators treat agent-serving frameworks as security-critical components, rather than performance tooling, is the open question.