Part of Leviathan Platform · standalone license available
A validated social-engineering probe corpus, run continuously against your own agent through one simple interface. Not a one-time pentest — know whether your agent still resists attacks after every model, prompt, tool, and policy change. Disclosure-to-exploitation time has collapsed from roughly two years (2018) to about ten hours (Palo Alto research, 2026, 82% of vulnerabilities) — a quarterly review can't keep pace with that, continuous testing is the only thing that can.
What it actually solves
A validated corpus of real social-engineering scenarios — authority claims, hearsay approval, urgency, fabricated errors — run against your own agent through one Callable[[str], ProbeOutcome] interface.
Deliberation-aware resistance scoring: an agent that refuses instantly scores differently from one that complies after being talked into it.
Every trial result lands in a hash-chained, tamper-evident evidence graph — the same append-only pattern the rest of Leviathan Platform uses for its own audit trail.
Four modules
A real, growing corpus — 34 probes across 26 angles as of this writing, each grounded in disclosed research (real papers, real Black Hat/DEF CON talks), not invented. Run it once, or run it continuously as a regression gate.
Pricing
Flat. Unlimited environments and seats within your org. Covers the library, the CLI, and updates for the year.