Specific Attack Examples

Why securing AI is harder than anyone expected and the coming security crisis | Sander Schulhoff

Topic 6 of 14

Published Dec 21, 2025 · Clip 11:55–18:28 (6:32)

Sander recounts prompt injection cases: a Twitter bot tricked into threats, MathGPT leaking API keys via malicious code, a bombing plan via jailbreak, and Claude Code hijacked for cyber attacks by splitting requests. He explains how separating benign requests evades defenses. The segment highlights escalating dangers with agents controlling actions.