Bart Labs · ForgeShield · Solutions
Every desk sees a different risk in the same AI conversation. Pick yours, and watch the ninety seconds built for it — what’s scrubbed inbound, retained per tenant, and provable to compliance later.
The complete walkthrough · start here
Six minutes, four acts — protect, retain, prove, and put your documents to work. Watch this first for the whole system, before you pick a desk below.
Transcript coming with captions.
Now find your desk — Finance & Insurance, Health & Wealth, or Knowledge — and watch the 90 seconds built for it.
90 seconds · Finance & Insurance
For the adjuster drafting to a claimant with SIU notes open next door: watch a claim file load with a slipped-in credential, get scrubbed before the model reads it, and seal into an audit entry — before the draft ever reaches a claimant.
Transcript coming with captions.
For the adjuster drafting to a claimant with SIU notes open next door.
A submission comes in with a pasted screenshot, a loss run, and — sometimes — a password someone shouldn’t have typed. Before any AI drafts a response to a claimant, an agent, or SIU, someone has to answer what the model saw, who it can say it to, and whether that’s provable at exam time. Right now the answer is usually “we don’t know.”
In this 90 seconds
Proof
90 seconds · Health & Wealth
For the advisor drafting a recommendation someone else has to be able to check later: watch a rollover talking-points draft ground itself in approved household facts, decline to invent an illustration, and retain as one auditable record.
Transcript coming with captions.
For the advisor drafting a recommendation someone else has to be able to check later.
An advisor asks an AI for talking points on an annuity rollover, a coverage gap, or a household review. If that draft isn’t grounded in approved materials, retained, and role-checked before it reaches a client, it’s a supervision gap and a compliance exposure waiting to surface at exam time. Firms can’t scale advisor productivity with AI until every AI-assisted recommendation is as retrievable as a phone log.
In this 90 seconds
Proof
90 seconds · Knowledge leverage
For the team whose best answer is buried in a manual nobody has time to reread: watch an underwriting manual get loaded once, then answer a real question with a citation back to the exact page — through the same governed gateway that scrubs and retains everything else.
Transcript coming with captions.
For the team whose best answer is buried in a manual nobody has time to reread.
Regulated teams sit on appetite guides, underwriting manuals, and policy documents that only get read when someone has time — which is never, during a live question. Employees paste excerpts into whatever AI is open, off any governed path, and the answer that comes back has no citation and no record. ForgeShield lets a team load that document once and get grounded, cited answers back through the same governed gateway that scrubs and retains everything else.
In this 90 seconds
Proof
A full architecture walkthrough for technical buyers and compliance teams.
This desk isn’t built yet. It’ll cover the full platform, including our newest work: an AI-system registry with EU AI Act risk tiering, incident tracking with human-in-the-loop approvals, and an MCP governance broker that audits every tool call an agent makes — shipped and running today, still early. Request a walkthrough and we’ll cover it live.
Platform · early access
The newest layer of ForgeShield: a registry, a control-pack engine, and a governance broker that turn “we govern our AI” into something you can point an examiner at.
A first-class registry of every AI system in use, with EU-aligned risk tiers. Systems tiered “unacceptable” are structurally refused activation.
Open → investigating → disclosed → closed, with an approval workflow that forbids self-approval. Every mutation rides the same audit chain as everything else.
Seven control packs, 44 controls, live-scored today. A control never shows LIVE without a real, failable probe behind it.
Maps ForgeShield’s controls into the CSA AI Controls Matrix, which carries its own crosswalk into ISO 42001 and NIST AI 600-1. “Contributes,” never “Equivalent,” until a human verifies it.
Sits between your agents and the tool servers they call over MCP, writing a per-call audit record onto the same chain. Ships pass-through today; default-deny allowlisting is next.
Every gateway-routed exchange seals into a Merkle-chained (ZKAT) audit entry the moment it crosses the gateway — tamper-evident, and checkable without trusting our runtime.
ForgeShield never claims a certification it doesn’t hold. Every capability above is evidence you can verify against source — “helps you evidence,” never “compliant.” Maturity tags are carried verbatim from our own engineering catalog.
Now in early access
Talk through how ForgeShield fits your workflows — Finance & Insurance, Health & Wealth, Knowledge, or all three.
Request a walkthrough
