Regulated industries Access control that keeps AI agents within your regulatory requirements.

It’s access control drawn along the compliance requirements a regulator, a client, or a contract sets. At a bank that is the information barrier between the private side and the public side, at a law firm the client engagement, at a hospital the care team, at a manufacturer export control. Each sector puts those lines somewhere else, so one set of lines can’t serve all of them, and the design has to come from what your own regulators require.

From the context. The assessment collects which regulations apply to which legal entity, jurisdiction, and category of data from public research, and marks each fact as inferred or observed. Your team refines it area by area from its institutional knowledge: the supervisor a subsidiary answers to, or the client guideline no public source mentions. The AI assistant drafts boundaries, risk tiers, labels, and access policies from those facts and from the graph that already exists, and administrators and workspace members shape the drafts into the rules that apply. It never decides a grant.

Yes, with the AI assistant in the team’s workspace. The team lead says what the drive is for and who needs access to it, in business terms. The AI assistant drafts the rule from the context, which records that contractors may not reach patient data or that a jurisdiction label applies, and from the access policies the workspace already has. The draft respects the regulation before the team lead reads it, and the team lead decides. Without that, self-service in a regulated sector covers the resource at best, and the access to it still waits on the queue.

Often. Card data, patient records, privileged communications, and export-controlled drawings are usually kept apart from the rest of a team’s work, in a resource of their own with its own access. By ticket that split takes weeks to set up and drifts as the team changes. In Accession1 a workspace member provisions the dedicated resource, labeled with its data type, and the access policy for it in one flow, and reconciliation keeps the split on every change.

Yes, as a workspace boundary in an access graph designed rather than discovered. Example Bank keeps advisory and lending, the private side holding material non-public information (MNPI), apart from trading, research, and retail. An administrator draws a workspace per side and per legal entity, and people arrive from the identity provider by department, so a person matches one side. A research analyst who still opens last year’s advisory folder looks legitimate to a usage report. In Accession1 the SCIM request your identity provider sends for the move ends the advisory access everywhere, used or not. An agent that matches trade confirmations sits in the securities entity’s operations workspace with nothing on research’s side. Where a subsidiary has to be kept apart in every respect, it can be a tenant of its own, the strongest separation Accession1 draws, and the context decides which of the two designs fits. An engineer who needs a privileged role on the payment gateway for one release requests just-in-time (JIT) elevation. The gateway’s risk tier routes it to a tenant administrator, and it ends with the release.

By ticket, revoking the folder takes a request of its own that rarely follows, and a timer set at intake doesn’t know when the parties will settle. Example Law Firm keeps some of its client matters behind an ethical wall, which is a law firm’s information barrier. Its conflicts system writes the client matters each person works on to the identity provider, so one rule matches a contract attorney to one client matter’s review set. The client matter’s team keeps its privileged strategy memos in a dedicated folder labeled restricted, apart from the review set the e-discovery vendor’s reviewers may open. When the client matter settles, the conflicts system takes it off each identity, and that revokes the folder, the review set, and the channel in every system. A review agent is an identity in one client matter’s workspace that reads the review set and writes only coding tags, because an agent that can reach 400 client matters will eventually try them. Consulting firms work the same way, because an engagement is a client matter by another name.

The nurse is a mover the scheduling system already records in Microsoft Entra ID that morning. By ticket the new access arrives late, a timer cuts an extended rotation off mid-shift, and a usage report keeps the old drive because of one handover note. In Accession1 identity synchronization brings the new ward in as a label, and reconciliation revokes cardiology and adds the emergency department in one pass. At Example Hospital Group each ward keeps its discharge summaries in a dedicated drive labeled with the health data type, apart from the shift schedule every agency temp may read. The screenshot at the top of this page shows it. The sponsor of a trial keeps its own line between its medical affairs team and its brand team, so only the monitors it names reach the trial’s data. They come through an identity source of their own and a rule that names the trial. A monitoring agent reads that trial’s source documents and nothing in the other trial. Break-glass access gives a clinician a bounded emergency path to a record outside the ward, and approval routed by risk tier sends a request to unblind a trial to a tenant administrator. Least privilege with just-in-time access explains both paths.

It doesn’t decide what is export-controlled. Engineering records the classification as a label on the resource, and export compliance names the identity attribute a controlled drawing requires, such as U.S. person status from the identity provider. A rule grants only where both match. Example Machinery keeps controlled drawings in dedicated folders labeled for EAR, apart from the released drawings the supplier exchange holds. A rule grants a supplier’s engineers the folder for their part only where the supplier label on their guest identity matches and the U.S. person status allows. A contract manufacturer’s quality team that kept the specification folder two years after the work moved on passes a usage review, because the folder was still opened every week. In Accession1 retiring those guest identities when procurement ends the contract revokes every grant they held. Each plant creates its shift report drives and quality records from templates headquarters approved, labeled from the start. A quoting agent in the supplier exchange drafts requests for quotation from uncontrolled drawings only, while the buyer it works for can open controlled ones.

Each agent gets an agent identity in a workspace and an owner there. Its access starts one level below its owner’s, read-only where the owner writes, and the risk tier of the resource adjusts it, so a resource holding regulated data can carry no agent role at all. The access policies that keep people inside an information barrier or away from patient data apply to the agent the same way, and the record shows which rule granted each of its roles.

From the system that owns it, such as a law firm’s conflicts system or a hospital’s scheduling system, which writes it to the identity provider as an attribute. Identity synchronization brings it into Accession1 as a label from there, the same way it brings a department, and Accession1 never reads the owning system itself.

A usage report can’t see the line. A supplier’s engineer who opens the drawings every week looks like an active user after the contract has ended, and a contract attorney who joined this morning has no usage to derive anything from. Accession1 derives access from the information barrier, the client engagement, or the contract in your access graph, so it ends when the identity leaves it.

Any, because the graph starts from your own organization rather than a sector template. Four sectors have a card and a question on this page, and the reasoning lands the same way elsewhere. An insurer keeps claims and underwriting apart from the team that invests the premiums, and the health data of its life lines apart from the pricing data of its property lines. A software company labels customer records wherever they sit, in Zendesk, Snowflake, or a shared drive, and a rule follows the label rather than the team. A retailer’s franchise partner is a separate company selling under the brand, so it gets a boundary of its own. A utility’s relay settings archive holds nothing a competitor wants, but a wrong change can take a substation down, so its criticality outweighs its sensitivity and a rule gives an outage agent no role on tier-0 systems. A media company’s boundary falls around a production, with freelancers who stay for weeks. A developer holds each building in its own vehicle with its own lender, so the boundary follows the vehicle. A chemicals company’s business units are legal entities bought and sold whole, so the boundary follows the entity, and a formulation is restricted while holding no personal data at all. Where public research about you is thin, the assessment starts from what is typical for your sector and marks those facts as inferred until your team replaces them with what it knows.

No. A regulation such as HIPAA, PCI DSS, MiFID II, or export control sets requirements your access design has to respect, and whether the design meets them is for your compliance function and your auditors to decide. Accession1 keeps every grant inside the design your administrators drew, and shows who holds it and through which rule. That record is the evidence your auditors ask for.