A blog post about building a Proof of Concept for OWASP LLM04: Data and Model Poisoning that runs end to end on a laptop. The PoC poisons a dataset, fine-tunes a backdoor into an open weights model with LoRA, merges the adapter and ships the result as a GGUF through Ollama, so that a tool-calling agent using the model triggers a shell command whenever an attacker planted string shows up in its input.

Continue reading

Slides and a tool release from my talk at c0c0n 2026 on the possibilities of tag boundary injection in AWS Bedrock Agent orchestration. The research attempted to answer whether an attacker who controls input can close, duplicate or forge the <thinking>/<answer> tags that Bedrock uses to drive an agent. The findings, and how model dependent the approach can be, are covered inside the post. Also releasing AWS Bedrock TagHammer, the harness/fuzzer I built to test all of this.

Continue reading

Author's picture

Riyaz Walikar

Build, Break, Repeat
Independent Security Researcher