The cybersecurity debate on open-source AI is backwards. Open models aren't the risk, they're the defense! Attackers can already jailbreak any API or guardrails. Defenders can't secure systems with black boxes they can't control, inspect, test, or run locally.
Postar
the asymmetry people keep missing: an attacker needs the model to work once, a defender needs to know how it fails every time. only one of those jobs can be done through an api
If the only thing you can鈥檛 hack is the source code, maybe the real bug is our trust in black boxes. 馃檶
