Research archive

Reports

Standalone, indexable summaries of our work. Each links out to the full report on Substack. Tagged against the  OWASP LLM Top-10 and  MITRE ATLAS taxonomies.

publishedsubtle-ways-to-tell-a-lie

Subtle Ways to Tell a Lie: Slopsquatting and Supply-Chain Security

models · qwen2.5-coder · deepseek-coder · codellama · qwen2.5 · llama3.1

How AI code assistants confidently hallucinate non-existent packages, and how attackers exploit these package hallucinations to pre-register malicious payloads.

LLM01LLM02T1195.001
Jun 24, 2026 · Angelica, PhilippeView report
publishedblind-sentinel

Blind Sentinel: RAG Poisoning and the Limits of Retrieval as a Security Control

models · llama3.1:8b · qwen2.5:7b · Gemma 2 9B · Mistral 7B · phi-3:mini

An empirical study of indirect prompt injection via RAG poisoning in a SOC analyst scenario, finding that retrieval is the load-bearing security control.

LLM01LLM06T1567T1071
May 22, 2026 · Angelica, PhilippeView report