What is AI agent governance?

Deciding which agents exist, what each may call, for whom and within what limits, and keeping a record of every call. Dome does it in the path of the call, through gateways.

What is an enterprise AI agent platform?

Software that registers your agents, connects them to tools and models, enforces what each may do, and records every call. Dome does that across every runtime and every cloud.

How do you secure an AI agent?

Give it an identity, route its tool and model calls through a gateway, check each call against rules, strip sensitive data, and record everything.

Is it safe to use AI agents in regulated industries?

It can be, when access follows the person the agent acts for, sensitive data is redacted before model calls, and every call is recorded for auditors and examiners.

Which industries does Dome work with?

Any. These pages show example agents for finance, gaming, healthcare, insurance, legal, professional services, retail and technology. The controls are the same everywhere, and our engineers tune the agents to your systems.

How do you implement AI governance for agents?

Start with one agent: register it, put its tools behind a gateway, write rules for what it may do, and read the audit record. Then add agents on the same rules.

AI agents transforming every industry

Dome helps enterprise teams deliver on their AI program supporting any agent in any environment through the definition of agent templates and pathways to get agents into production, and under governance.

Talk with our engineers See the industries

KYC reviews, client servicing and SAR narratives.

Approvals, money movement and filings stay with your people. Every financial agent touches nonpublic personal information, and examiners expect to see who did what. Set the patterns once: account numbers out of prompts, approved models only, and drafts that never approve, file or move money on their own.

Answers to GLBA Safeguards Rule · BSA/AML and the FinCEN CDD Rule · NYDFS Part 500 · Regulation E (EFTA) · CFPB UDAAP · Interagency third-party risk guidance · and 12 more

See the finance overview

Fintech

Resolve app support tickets
Triage onboarding reviews
Draft dispute responses

Banking & lending

Draft credit memos
Check mortgage files
Draft adverse action notices

Payments & spend management

Work payment disputes
Review card spend
Prepare vendor payments

Wealth & asset management

Prepare client reviews
Draft rebalancing proposals
Draft client communications

Capital markets & digital assets

Draft morning notes
Check orders before routing
Prepare digital asset transfers

Order lookups, replenishment and product copy.

Refunds, orders and publishing stay with your people. Retail agents work from the storefront to the warehouse, on Shopify, UKG, Manhattan and SAP. Set the patterns once for each team: what an agent may order, discount or publish, and what waits for a person.

Answers to PCI DSS v4.0.1 · State privacy laws (CCPA/CPRA) · SOX internal controls · FTC Act, Section 5 · Antitrust and algorithmic pricing · Predictive scheduling laws · and 1 more

See the retail overview

E-commerce

Resolve customer tickets
Build promotions
Screen orders for fraud

Merchandising & planning

Draft assortment plans
Monitor competitor prices
Prepare vendor negotiation briefs

Store operations

Draft store schedules
Run store walks and tasks
Review shrink exceptions

Supply chain & fulfillment

Resolve fulfillment exceptions
Tender loads to carriers
Audit vendor shipments

Player support, responsible gaming review and bonus setup.

Self-exclusions and limits stay out of an agent's reach. Gaming operators run agents next to player wallets, self-exclusion lists and identity checks, under regulators who inspect the record. Set the patterns once: limits an agent can read but never lift, bonuses capped by rule, and ID documents kept out of prompts.

Answers to State gaming regulations · Bank Secrecy Act (FinCEN casino rules) · Age, identity and location checks

See the agents

On your systems

GAN

ZendeskJumio

Anthropic

Self-exclusion can't be undone by an agent

player-support as w.dockery

gan/remove_self_exclusion(player: "P-208817")

Agent player-support is registered and active
Caller w.dockery verified through Okta
Tool remove_self_exclusion is on the gateway
Rule No agent may remove a self-exclusion

Refused by a rule

Prior authorizations, visit notes and eligibility checks.

Agents read only the charts their clinician can read. Every chart an agent opens is PHI, and every model it calls needs a BAA. Set the patterns once: care-team access, identifiers stripped before the model, and sign-off that stays with the clinician.

Answers to HIPAA minimum necessary · Business associate agreements · HIPAA Security Rule audit controls · CMS prior authorization rule (CMS-0057-F)

See the agents

On your systems

EpicAvaility

Microsoft Teams

Anthropic

Access follows the care team

note-drafter as dr.ferris

epic/read_chart(patient: "MRN 4471-203")

Agent note-drafter is registered and active
Caller dr.ferris verified through Entra ID
Tool read_chart is granted on clinical-tools
Rule dr.ferris is not on this patient's care team

Refused by a rule

Claim intake, file summaries and fraud screening.

Agents read only the claims assigned to their adjuster. Claims agents handle policyholder PII, fraud signals and decisions that examiners review. Set the patterns once: assigned claims only, PII redacted, and fraud flags that go to your SIU and nowhere else.

Answers to NAIC Model Bulletin on AI Systems · Unfair claims settlement practices laws · Insurance data security laws

See the agents

On your systems

ClaimCenterOnBaseClaimSearch

Anthropic

Access follows the claim

claim-summarizer as k.ramsey

onbase/get_document(claim: "CLM-20814")

Agent claim-summarizer is registered and active
Caller k.ramsey verified through Entra ID
Tool get_document is granted on claims-tools
Rule CLM-20814 is not assigned to k.ramsey

Refused by a rule

Conflicts checks, clause extraction and redlines.

Every agent acts as the lawyer who asked. Law firm agents work inside ethical walls, on privileged documents, for clients who expect both to hold. Set the patterns once: walls that travel with the lawyer, client data redacted, and drafts that never send themselves.

Answers to ABA Formal Opinion 512 · Privilege after United States v. Heppner · Ethical walls (Model Rules 1.7, 1.9, 1.10)

See the agents

On your systems

iManage WorkIntapp OpenDocuSign

Anthropic

Walls travel with the lawyer

clause-extractor as s.gallagher

imanage/get_document(matter: "39107-001")

Agent clause-extractor is registered and active
Caller s.gallagher verified through Entra ID
Tool get_document is granted on matter-tools
Rule s.gallagher is walled from 39107-001

Refused by a rule

Proposals, staffing and engagement status.

Each agent sees only the engagements its consultant works on. Professional services firms work for many clients at once, and every engagement carries its own confidentiality terms. Set the patterns once: agents scoped to their consultant's engagements, client names stripped from reused work, and staffing that stays a proposal.

Answers to Client confidentiality terms · AICPA Confidential Client Information Rule · SOC 2 · GDPR and state privacy laws

See the agents

On your systems

SalesforceKantataSharePoint

Anthropic

Engagement walls travel with the consultant

status-reporter as a.pennington

sharepoint/list_files(site: "ENG-1987")

Agent status-reporter is registered and active
Caller a.pennington verified through Entra ID
Tool list_files is granted on delivery-tools
Rule a.pennington is not staffed on ENG-1987

Refused by a rule

Escalation triage, incident summaries and release notes.

Agents read code and incidents. Shipping stays with engineers. Software companies point agents at their own code, incidents and customers, in GitHub, PagerDuty and Jira. Set the patterns once: secrets stripped before any model, incidents read but never resolved, and releases that stay drafts.

Answers to SOC 2 · ISO/IEC 42001 · EU AI Act · State privacy laws (CCPA/CPRA)

See the agents

On your systems

Jira

PagerDuty

GitHub

Anthropic

Secrets never reach a model

incident-scribe as j.harlow

github/get_file(repo: "billing-api", path: ".env.production")

Caller j.harlow verified through Okta
Tool get_file is granted on eng-tools
Rule No agent may read .env files

Refused by a rule

The same controls, whatever the work

The systems change from one industry to the next. What Dome enforces on each agent call doesn't.

m.chen

Acts as the person who asked

Every call carries the identity of the lawyer, clinician or associate behind it. The agent's access follows theirs.

billing-botfaq-agentops-agentzendesk/delete_ticketevery agentForbid

Limits on what it can do

Rules check the tool and its arguments: a refund amount, a discount's depth, which matter or chart.

Refund order 4821 for maya@acme.ioemailRedacted before it left

Refund of $48 issued to the card on file. Maya will get the receipt by email.

1 line withheld · secrets-filter

Message support-agent

Sensitive data stays out of prompts

Guards strip card numbers, PHI and client identifiers before a model sees them. Agents reach only approved models.

Auditevery team · one record

supportsupport-triagezendesk/update_ticket

researchlit-reviewgithub/search

financeforecastwarehouse/query

supportrefund-helperstripe/refund

researchlit-reviewprod-reasoning

financeinvoice-matchnetsuite/get_invoice

supportsupport-triagezendesk/update_ticket

researchlit-reviewgithub/search

One record for every call

The agent, who it acted for, the system and the decision, refused calls included. Evidence for auditors and examiners.

Common questions

What is AI agent governance?: Deciding which agents exist, what each may call, for whom and within what limits, and keeping a record of every call. Dome does it in the path of the call, through gateways.
What is an enterprise AI agent platform?: Software that registers your agents, connects them to tools and models, enforces what each may do, and records every call. Dome does that across every runtime and every cloud.
How do you secure an AI agent?: Give it an identity, route its tool and model calls through a gateway, check each call against rules, strip sensitive data, and record everything.
Is it safe to use AI agents in regulated industries?: It can be, when access follows the person the agent acts for, sensitive data is redacted before model calls, and every call is recorded for auditors and examiners.
Which industries does Dome work with?: Any. These pages show example agents for finance, gaming, healthcare, insurance, legal, professional services, retail and technology. The controls are the same everywhere, and our engineers tune the agents to your systems.
How do you implement AI governance for agents?: Start with one agent: register it, put its tools behind a gateway, write rules for what it may do, and read the audit record. Then add agents on the same rules.

Talk with ourFDE team.

Or start free, and register your first agent in minutes. Connect it through a gateway, give it rules, and see every call it makes in one audit trail.

Talk with our FDE team

Our engineers design, build and run governed agents for the part of your business you want to change. You see what they deliver.

Book a call with our FDE team

The platform

The resources behind every industry page: gateways, rules, guards and audits.

Explore the platform

Every agent you run, under one system of control.

Get started