Jailbreak gate
Heuristic detection of instruction overrides, DAN-style prompts, and system-prompt exfil attempts.
Closed source · AI safety
Bastion
Closed-source safety library and studio: score prompts for jailbreaks, apply ethical policy packs, and gate risky model output before it ships.
Get the free desktop studioCapabilities
Ethics packs
Harm, deception, privacy, dual-use, and child-safety guidance baked into decisions.
In + out
Score the inbound prompt and the outbound draft — defence in depth for agents.
Open companion: Open Bastion — MIT subset for prompt gating.
Download
Enter your email to unlock Bastion.
Free. Closed source. We’ll notify SanRa when you download.