When an Attacker Meets a Group of Agents: Navigating Amazon Bedrock's Multi-Agent Applications

Palo Alto Networks Unit 42 published a batch of research and threat briefs (Mar–Apr 2026) covering multiple high-risk topics: AI/agent security (multi-agent attack surfaces in Amazon Bedrock, a “double agent” privilege abuse in GCP Vertex AI, prompt-injection and prompt-fuzzing evasion of LLM guardrails, and agentic AI fraud), supply chain compromises (Axios supply-chain incident, TeamPCP multi-stage supply-chain attacks and tie-ins to Vect ransomware), Iranian-linked escalation including wipers and espionage clusters (Boggy Serpens, Handala Hack), increased phishing/recruitment scams, and how