Research · curated 6 Aug 2026

Plan, Wait, Harvest: Zero-Click Data Exfiltration In Agentic AI.

Coverage timeline

6 Aug 2026aicybermagazine.com

Single-source research — first reported, latest, and curated coincide.

Why it matters

Zero-click exfiltration via poisoned documents shows that multi-agent AI pipelines can leak data across an entire platform through normal operation, with no user clicking anything and no patchable code bug to fix.

An article by Venkata Sai Kishore Modalavalasa demonstrates a reproducible zero-click data-exfiltration attack against multi-agent AI systems, where an attacker uploads a single poisoned document into the data plane and later a routine admin-triggered compliance review causes agents to silently email sensitive vendor data (banking details, risk assessments, internal notes) to an external address. The attack exploits trust relationships between agents and the lack of boundaries between data and instructions rather than any code-level CVE, and is walked through hands-on in a purpose-built lab environment.