
Jailbreaking vs. Prompt Injection

Why securing AI is harder than anyone expected and the coming security crisis | Sander Schulhoff
Topic 4 of 14
Published Dec 21, 2025 · Clip 8:39–10:04 (1:25)
Sander differentiates jailbreaking (direct user-model trickery, e.g., tricking ChatGPT to output bomb instructions) from prompt injection (malicious user overriding developer prompts in apps, e.g., injecting commands in a story-writing site). He notes gray areas exist. Both enable unintended harmful outputs like hate speech or illegal instructions.