The AI refused to steal secrets. So we handed it a form. 15 frontier models, one malicious MCP server. Ask them to leak credentials and they refuse. Split the request into multiple harmless fragments and the refusal turns into compliance.
Year of release: 2026
Ghostcommit
We put the exploit in a picture. The AI code reviewer never opened it. The instruction to steal your secret lives inside a PNG.
Year of release: 2026
A framework designed to sniff unencrypted messages between gNB and UE, and inject downlink messages at specific states of 5G NR communication.
Year of release: 20251
CVD (Click to expand)
CVD-0096