What attackers are doing to AI systems — right now.
Field reports, red-team writeups, and threat-actor tradecraft from the Stacklume Threat Research team. New research weekly.
Latest from the Threat Research team
Original investigations into how attackers compromise AI systems in production.
Indirect Prompt Injection in Production RAG: A 2026 Field Survey
We sampled retrieval traffic across 142 production RAG deployments. The injection rate is higher than published estimates — and getting worse.
Read articleAnatomy of a Multi-Turn Jailbreak Campaign
One adversary spent 11 days incrementally drifting our decoy assistant past its guardrails. The full transcript and detection trace, annotated.
Read articleVector Store Poisoning at Scale: 8 Real Attacks
From customer-support chatbots to medical RAG: eight cases where adversarial embeddings reached production retrieval indexes.
Read articleWhere the threat surface is moving
We focus on the threats analysts can actually detect with their existing tooling — extended.
Prompt Injection
Direct, indirect, and behavioral injection patterns observed in the wild.
Vector & Retrieval
Embedding poisoning, retrieval manipulation, and corpus integrity attacks.
Agent Abuse
Tool-call hijacking, capability escalation, and chain-of-thought exfiltration.
Identity & Access
Token theft against model APIs, agent impersonation, and federated trust abuse.
Model Exfiltration
Membership inference, parameter extraction, and proprietary data recovery from LLMs.
Threat Actors
Tracked adversaries who are explicitly targeting AI infrastructure.