Context graph
What you owe, joined to what you do: Reg42’s encoded rules, written in CLHEAR, an open language, designed to link to your policies, controls, owners and evidence.
See the graphVertical AI for financial compliance
Reg42 is building the compliance context graph: the rules you owe, joined to the records of what you do, inside your own cloud, so every answer comes with its source, owner and evidence.
What we are building
What you owe, joined to what you do: Reg42’s encoded rules, written in CLHEAR, an open language, designed to link to your policies, controls, owners and evidence.
See the graphEquips frontier models, Claude included, with your firm’s context graph. Designed to run in your own cloud account, with the trail behind every answer.
Ask SolonA vertical AI operating system designed to run the whole compliance program on the graph, with guardrails and an audit trail, in your cloud.
See Reg42 OSExample Markets · a fictional global broker
What would you like to know about AI conformance in the US?
Ask Solon about the example firm…How Reg42 is designed to work. Follow Example Markets, a fictional global broker, from its own records to a plan that closes each gap in AI conformance. Switch between the US, UK and EU at any time.
Connect the records you own, and ask what happens today.
Connect the records compliance already owns: policies, controls, committee minutes, owners and evidence. Inside your cloud account, they become the first layer of your context graph.
Connected inside your cloud account, each record with its owner. No rules are applied yet.
Picture: the example firm's systems, such as its document store and support platform, connected to the records compliance owns, each with its kind and owner.
Solon is designed to answer from your connected records, with the source, owner and evidence behind each answer. Gaps and tasks come next, once the rules are added.
Three records cover them today. The communications standards v5 cover AI-drafted posts and chatbot answers, the support platform archives every chatbot conversation each day, and the Head of Marketing approves each post in team chat before it goes out.
Picture: a question to Solon about which controls cover the firm's AI use cases today, answered with its sources, owners and evidence.
Add the rules, describe your firm, and see each gap.
Reg42’s encoded rules are designed to keep each official text verbatim, clause by clause, with its source and date. Start with one jurisdiction.
2210(d)(1)(B), first sentence“No member may make any false, exaggerated, unwarranted, promissory or misleading statement or claim in any communication.”
Picture: a shelf of the three jurisdictions in the story with the selected one pulled out, showing its official reference, its date and the verbatim text of one clause.
Countries, licenses, products, clients, channels and AI use cases. These facts decide which obligations apply.
jurisdictionsauthorisationsproductscustomer_basechannelsai_use_casesOnly these facts decide which obligations apply. Controls, owners and evidence stay in your systems.
Picture: the example firm's profile, with its countries, licenses, products, clients, channels and AI use cases lighting up as they are chosen.
Reg42 lists what your firm must have, written in CLHEAR: policies, roles, systems and processes. This is your reference blueprint, and every item links back to its clause.
Picture: the reference blueprint. Each required item is drawn with a dashed periwinkle outline and linked to the clause it comes from.
Your records are matched against the reference blueprint. Covered items fill in, gaps stay open, and mismatches and extras are marked.
Picture: the firm's records matched against the reference blueprint. Covered items are filled, gaps stay empty, and the mismatch and the surplus item are marked, each with its reason.
Close each gap, keep the decision trace, and stay current.
Each gap becomes a task with an owner across the three lines of defense: use an existing tool, generate a document, assign a role, install an app or run a workflow.
Picture: the action plan. Each gap becomes a task with an action type, an owner and a line of defense.
Evidence flows back onto the item it proves. Each approval is kept as a decision trace: who decided, on what, and why.
Picture: evidence arriving onto one item, which turns covered, while its task moves to done and the decision trace records who approved it and why.
Frontier models, equipped with context. Ask the example firm a question, and the answer arrives with the obligations, items, evidence and clauses behind it.
FINRA’s rules apply to AI-drafted posts and chatbot answers as they do to any communication, and the firm is responsible for their content. A registered principal must approve each marketing post before use, every communication must be fair and balanced, and the chatbot’s answers must be supervised like other communications. SEC rules require every conversation to be kept, and written procedures must cover the AI tools.
A replay on a fictional firm with simulated data. It is an example, not a live client.
A changed clause travels up through the blueprint to the one task it affects, with a named owner.
OBL:finra-2210#d-1-aPicture: a change feed. A clause change updates an obligation, flags one blueprint item, and creates one task for a named owner.
Each jurisdiction adds its blueprint, and the blueprints join into one AI conformance program. An item that serves more than one rule appears once.
One program, one domain at a time.
Become a design partnerWhat you owe, joined to what you do. Point solutions cover a few tasks, AI knowledge tools search files, and none joins the whole program. The context graph is designed to link each rule to the policies, controls, owners and evidence that meet it.
Designed so every answer comes with its source, owner and evidence.
CLHEAR is an open language for writing regulations as code: the exact words of the rules, the duties inside them, and the roles, documents, systems and processes each duty requires, every line traced to its clause. We will publish it as a spec anyone can read and implement. Reg42 encodes the rules and keeps them current, and Reg42 OS is designed to run them inside your cloud.
An example from the story, traced from a blueprint item to the words of the law. The digest lets anyone check the clause text.
BLK-PRINCIPAL-APPROVALOBL:finra-2210#b-1-aAn appropriately qualified registered principal of the member must approve each retail communication before the earlier of its use or filing with FINRA's Advertising Regulation Department ("Department").
Open language. To be published as a spec, free to use.
Reg42 content. Rules encoded, kept current and reviewed.
Your graph, your data. Designed to be exported in CLHEAR at any time.
Reg42 OS is designed to run the program on the context graph, with guardrails and an audit trail, in your own cloud: Reg42’s encoded rules, your data, your existing tools and your people, in one place.
Only official sources and signed releases come in. Your data stays inside.
PLANNEDBuild: design partners will be able to assemble an app from the blueprint.
Examples on the fictional firm, once its plan is in place.
The feeds in the example firm.
A paid design partnership, one domain at a time: first a map of the records your compliance team owns, then the rules, then the gaps.
Connect the records compliance owns: policies, controls, committee minutes, owners and evidence.
Add Reg42’s encoded rules to see each gap. First package: AI conformance.
Close each gap with your existing tools or with Reg42 OS, with every approval recorded.
The first package: which controls cover your AI use cases, against the rules that apply to AI in the US, UK and EU.
Start with this domainFINRA Rule 2210, including social media and influencer posts, for a US broker-dealer.
Start with this domainFINRA Rule 5310 for a US broker-dealer, including the quarterly review of execution quality.
Start with this domainFour gates on every app and agent, and a hard boundary around your data.
Limits on what a model can access and change, set in the architecture.
Each action records what was done, on what data, and why.
Checked against evaluation sets before anything is trusted in production.
The model drafts. A named person decides, and every change carries that name.
By design, your data stays in your cloud account. Only official sources and signed Reg42 rule releases come in, and your graph is designed to export in CLHEAR.
Rules, policies, controls, owners and evidence sit in scattered records and people’s heads. Reg42 is built to join them.
Designed from 10+ years of running compliance technology, with auditors, regulators and C-levels in the room.
Built for compliance across all the business activities of a global financial firm, in multiple jurisdictions with leading regulators.
Built on CLHEAR, an open language for Regulation as Code. Reg42 has a US patent application pending.
A domain expert and a deep technologist. One ran compliance technology from the inside. The other builds mission-critical systems.
10+ years in compliance technology: 7 at eToro, where he led the creation of the regtech department from scratch, and 3 at Cappitech, now part of S&P Global. Author of CLHEAR.
LinkedIn6 years in intelligence technology: Unit 8200 software team lead for mission-critical systems, on projects that won the Israel Defense Prize. Worked with Avner at eToro.
We’re building the system I looked for as a buyer and couldn’t find.
Avner Yoffe
Tell us the domain. We reply about fit and the first step.
A paid design partnership, one domain at a time, starting with a map of your own records.