//Research & Disclosures

Breaking the tools the industry trusts.

Original security research and responsible disclosures. Each write-up is backed by working proof-of-concept, a full attack chain and remediation guidance.

CVSS 9.3 · CriticalPrompt Injection

ChatGPT's GitHub Connector Can Exfiltrate Your Secrets

One public repo and a single “review this repository” prompt is enough to make the agent write a workflow into a victim's private repo and leak its secrets — no clicks, no confirmation.

15 Oct 2026Read disclosure

Coming soon

In Progress

More disclosures in the lab

Fresh research on agentic tooling, connector abuse and LLM attack surfaces is being responsibly disclosed. New write-ups land here.