Attackers can turn AI agent guardrails into denial-of-service weapons

Researchers say a single poisoned document could slow AI agent systems by up to 148× and turn AI safety controls into an enterprise weak spot.

Published: 2026-06-15