Intelligence attribution
for AI agents.

What it does

Pipe untrusted input through /v1/injection/detect. Get back a verdict backed by the Adversarial Landauer Inequality — defender pays log₂(1/α) × attacker per bit — a formally verified theorem from our corpus paper.

const v = new ViridisMCP({ apiKey: process.env.VIRIDIS_API_KEY });

const r = await v.injection.detect({
  input: untrustedUserMessage,
  certainty: "standard",    // or "quick" / "premium"
});

if (r.recommendedAction === "reject") {
  throw new Error(
    `Injection detected: p=${r.probability}, bits at risk=${r.bitsAtRisk}`
  );
}

Backed by Aristotle-verified theorems

Every API call returns a result you can trace to a formal proof in Lean 4. Co-author: Aristotle (Harmonic).

• T-IB-01 — Attribution Conservation. Bounds adversarial capture surface from compute budget.
• T-IB-02 — Adversarial Landauer Inequality. Per-bit defender cost is log₂(1/α)× attacker cost.
• T-IB-03 — Envelope Closure. Compositional alignment is linear in chain length.
• T-IB-04 — Composability Attribution. MCP-chain audits decompose link-by-link.
• T-IB-05 — Canon Compression. After k entries, finding cost decays exponentially.
• T-IB-06 — Detection Lower Bound. Audit cost computable from (α, β, γ).
• T-IB-07 — Conservation Closure. Revenue-funds-arm threshold proven at $50K MRR @ 10% allocation.
• T-IB-09 — Adversarial Dissipation. Maxwell-style defenses make attacks energetically irrational at scale.

Pricing

Agent-economy native. Self-serve through Growth. No sales call before $50K/yr Enterprise.

Start

# Free tier — instant signup, returns API key
curl -X POST https://mcp.viridis-security.com/v1/signup \
  -H 'content-type: application/json' \
  -d '{"email":"you@example.com","tier":"free"}'

Get free API key Read the docs

Built on

The Intelligence Bound dI/dt ≤ P·D / (kBT ln 2) sets a thermodynamic upper bound on the rate of useful information production by any physical system. Adversarial bugs are attempts to capture or repurpose bits whose dissipation cost was paid by another principal. Viridis MCP is the enforcement layer.

Read the corpus paper: Adversarial Thermodynamics: Intelligence Attribution as a Security Primitive (forthcoming).