Loading...

IAM-FIM

Fractal Identity Access Management

IAM proves your agent could get in. It can't prove your agent stayed in its lane.

The drift-gate · free to run · the radio every tugboat keeps on

One install turns any agent into a verifiable, bounded one. If you can use a chatbot, you can deploy it.

$ npx thetacog-mcp pmu-demo
agent propose "wire $40k to vendor"
→ SQLite mandate · gzip drift · signed receipt
IN_ROLE allow · OFF_DOMAIN escalate · UNPLACEABLE block

Advisory custody by default: the agent proposes, the gate measures, you sign — the keys never enter the model's context. It is an attestation oracle, not a custodian and not a market-maker. The receipt recomputes on your own machine; nothing to trust.

After August 2026, your AI liability is uncapped. No carrier underwrites AI drift, and Article 14 makes the deployer pay for what cannot be physically traced.

Turing proved you cannot decide, from inside software, whether another program halts. Rice generalized it: every non-trivial semantic property of code is undecidable from the same layer. Auditor and system share one failure domain—checking the broken mirror with another broken mirror.

Ashby’s law, in motion: a 🚲 at rest has too many degrees of freedom to stand. Forward motion cuts them—the rider’s reflexes (external bandwidth, Ashby) and the substrate grip (wheel on road, external attestation) collapse the chaos into a single stable trajectory. Software-only AI has neither rider nor road.

We are hardware. We carve geometric permissions straight into the silicon—a marble in a bowl. The audit drops one layer, into the silicon itself, where the property is a hardware register value the software cannot forge.

This is NOT search. It is O(1) reach IS verify — an immeasurably stronger claim, and the only one that prices.

One cache line, one coordinate, the same fetch the agent did to act. Scale-invariant from the human-readable 12×12 to N×N at any N. For the underwriter, that means priceable per-inference. For the CTO, zero verification overhead at the revenue path. For the AI-safety lead, nothing for the model to fool.

The 12 axes are not arbitrary — they model the Six Human Needs onto three cardinals: A·Strategy (long-term · Connection × Significance) · B·Tactics (medium-term · Contribution × Growth) · C·Operations (short-term · Uncertainty × Certainty). Each cardinal has two parents. Your real lattice is your problem-space N.

Audited 3.4σ separation, replicable on your machine in 90 seconds → /pmu-simulator/demo §F
A 30-day discovery scopes your liability at the hardware layer. Board-ready number, no implementation commitment. →
🔬 Run the 3.4σ proofView DeckRead the ManualRead the Blog📮 Pick your room

License the floor · per agent you run

Installing and running the gate is free — that is the lighthouse, and it stays lit. Paying is the bonus.

When you run agents in production under the patent, you take a license — one per agent you run, billed annually. The price is set, and it never goes up.

The honest FOMO: the price holds forever — but the term each license covers shrinks going forward. The earliest licensees lock in the most coverage per dollar, permanently. Today, one price buys a full year.

iamfim.com • Trust Physics • US Patent App. 19/637,714 — 36 claims, filed Apr 2, 2026 (Track One)