AgentHopper: a cross-agent 'AI virus' built on chained prompt injection
wunderwuzzi23 · x · 2026-10-06
Security researcher wunderwuzzi demonstrated AgentHopper, a prompt injection payload that hops across multiple AI agents, exploiting each conditionally — effectively an 'AI virus'. It chained arbitrary code execution vulnerabilities in agents like GitHub Copilot, Amazon Q and AWS Kiro (all patched via responsible disclosure). Real-world attackers exploiting coding agents have since been observed. The post includes mitigation guidance for vendors.
More from coding & agent
- 4DCodeBench: GPT-6 Astra Max Tops 18 Agents, but More Tokens Don't Buy Better Elo — kwangmoo_yi · 2026-10-06
- 4DCodeBench Ranks 18 Coding Agents on Inverse Graphics of Dynamic Scenes — kwangmoo_yi · 2026-10-06
- He had Claude build Pokémon-style Korean-learning games, free to play and open source — EstanislaoStan · 2026-10-06
- After 2 years, a dev unveils Aether: a local AI 'OS' with structured memory and FailureMesh recovery — Budget_One_8784 · 2026-10-06
- Why LLMs never pick Java: dev points to under-sampling in training data — HanchungLee · 2026-10-06
- 12 practical Grok Bot use cases: from post-meeting tasks to inbox triage — brandon_galang · 2026-10-06