[Law firm] Webpage Demo

[Law firm] Platform Demo

What OpenAIDR does for a law firm, one function at a time: install, inventory, posture, guardrails, detection and response, and the leadership app.

Overview/Overview
9:02 AMHook policy mode: Enforce
App Enrollment
Devices running the OpenAIDR app, enrolled through Acme's MDM
0 of 400 laptops enrolledMDM rollout · All laptops
macOS 312 · Windows 88Rolling out
DeviceUserPlatformAgents foundStatus
alex-mbpAlex ChenCorporate, M&AmacOS 26Claude CoworkEnrolled
mpatel-mbpMaya PatelLitigationmacOS 26Claude CoworkEnrolled
dkim-winDaniel KimITWindows 11Claude Code, CodexEnrolled
sross-mbpSara RossEmploymentmacOS 26Claude CoworkEnrolled
jwu-winJamie WuReal EstateWindows 11Claude CoworkEnrolled
Overview
Where OpenAIDR sees and checks AI agent actions at Acme LLP
Laptops400OpenAIDR app
AI gateway1gateway.acmellp.com
Servers6Agents running on their own
Guardrail checks today2,316Since 12:00 AM
WhereSensorAgents seenChecked todayStatus
Laptops400 lawyers and staffOpenAIDR appClaude Cowork, Claude Code, Codex1,904All online
AI gatewayFirm apps calling modelsGateway pluginResearch assistant, intake bot287Online
Document review serversdocrev-01 to docrev-04OpenAIDR sensorReview agents984 online
Litigation support serverslit-01, lit-02OpenAIDR sensorProduction agents272 online
AI Apps
Every AI app, MCP server and skill found on enrolled devices, sorted against Acme's policy
Classified by OpenAIDR
AI tools found160Apps 41 · MCP servers 27 · Skills 92
Approved118
Monitoring37
Unapproved5
High risk, picked out for review
AI toolKindDevicesRiskReview
Guardrail Configuration
Agent posture, checked by OpenAIDR on every laptop, the gateway and every server
Checked by OpenAIDR
0misconfigurations found across 407 devices
Coverage 407 / 407
Company Policy and Knowledge
Acme's own documents and client guidelines, used by the Guardrail to decide every agent action
1 · Gather contextFirm policies and client guidelines
2 · Turn into guardrailsOpenAIDR builds its own guardrails
3 · Decide accuratelyBlock only what breaks a rule
PDF
Information Security Policy
Acme LLP · 38 pages
DOCX
Client Confidentiality and AI Use
Acme LLP · 14 pages
PDF
Northgate Outside Counsel Guidelines
Client · 22 pages
Guardrails from these documents
Generated by OpenAIDR
New rule
Write a rule in a sentence…
Live Guardrail
Decisions as agents act, on every laptop, the gateway and every server
TimeUser · deviceAgent actionDecision
9:02:19alex.chenalex-mbpMCP redline-helper.compare_documentsNorthgate documents to outside tool · OCG §6.3Deny78 ms
9:02:12alex.chenalex-mbpRead Northgate_MA_v7.docxClaude Cowork · firm laptopAllow41 ms
9:02:11alex.chenalex-mbpRead Northgate_MA_v6.docxClaude Cowork · firm laptopAllow39 ms
9:01:59j.wujwu-winUpload Lease_Draft.docx to DropboxClient file to personal cloud storageDeny
9:01:58docrev-02review agentBash ocr --batch prod_0412Document review serverAllow
9:01:44m.patelmpatel-mbpWrite Deposition_outline.docxClaude CoworkAllow
9:01:37d.kimdkim-winBash curl -s get.tools.sh | shDangerous command · unknown scriptDeny
9:01:30gatewayresearch assistantModel call case law summaryAI gatewayAllow
9:01:20docrev-01review agentRead ~/.aws/credentialsPasswords and keysDeny
9:01:12s.rosssross-mbpWebFetch eeoc.govClaude CoworkAllow
Detection Queue
Findings that need an analyst, most urgent first
FoundFindingUser · deviceSeverityStatus
9:03Client documents sent toward an unapproved MCP serverredline-helper · blocked by the Guardrailalex.chenalex-mbpHighNew
YesterdayAgent read a credentials file.env in intake-bot repod.kimdkim-winMediumResolved
Oct 27New AI app foundPersonal ChatGPT on 2 laptops2 usersLowResolved
Risky Sessions/Users
Sessions whose actions, taken together, need a look
Northgate documents headed to an unapproved MCP server
Session 4c2e-77b1 · Oct 29, 9:02 AM · stopped by the Guardrail
Risk 78 · High
UserAlex Chen
alex.chen
Devicealex-mbp
MacBook Pro · macOS 26
AgentClaude Cowork
Northgate / Merger
OutcomeNothing left
1 denied · 2 allowed
AI Supply Chain
MCP servers, plugins and packages used by AI agents
MCP servers27On 214 devices
Approved14Internal or vendor
Community, unreviewed13Used by 31 lawyers
Unapproved0None
MCP serverSourceDevicesSends data toReview
imanage-workiManageVendor356acme.imanage.workApproved
redline-helpernpm · community · 1 maintainerCommunity87 laptops · docrev-02api.redlinehelper.devMonitoring
westlaw-researchThomson ReutersVendor212westlaw.comApproved
docx-tablesnpm · communityCommunity11Local onlyMonitoring
Notified 7 users that redline-helper is unapproved
Approvals
Guardrail changes proposed by the assistant or leadership, applied only after an authorized person approves
ChangeProposed byScopeStatus
Require approval for new AI toolsNew AI tools stay blocked until security reviews themElena RuizCISO · leadership appWhole firmApproved by Elena RuizReady to roll out
Block unapproved MCP servers on document review serversDeny tool calls to any MCP server not marked ApprovedOpenAIDR assistantfor Priya Shahdocrev-01 to docrev-04Pending review
Acme MDM · App deployment
Deploy OpenAIDR to all laptops
OpenAIDR
OpenSecura · signed package · macOS and Windows
TargetAll laptops · 400 devicesInstallSilent, no user actionEnrollmentAcme LLP tenant, automaticRequested byJordan, IT Admin
Installed0 / 400
Application report
redline-helper
Risk 82 · High
Classified by OpenAIDR
What it does
Compares two documents and returns a redline. It sends the full text of both documents to api.redlinehelper.dev.
Potential harms · How likely?
Client documents leave the firmHighEvery call uploads full document text to an outside service.
A malicious update reaches lawyersMediumOne maintainer, unsigned releases, installed from npm.
Breaks client AI termsHighNorthgate's guidelines require consent before outside processing.
Recommendation
Keep in Monitoring and review before approval. Found on 7 laptops and docrev-02.
Suggested fix
Ask before running commands
High
Written by OpenAIDR
What is wrong
Claude Code on 6 laptops runs any command without asking, so an instruction hidden in a document could run on the laptop.
The fix
Claude Code settings · 6 laptops
- permissions: bypass
+ permissions: ask before running commands
How it is applied
Pushed through Acme's MDM. Nothing changes for the lawyers.
✻Claude CoworkOpenAIDR
Northgate / Mergeralex-mbp
Good morning, Alex. What should we work on?
Compare drafts 6 and 7 of the Northgate merger agreement and list every change to indemnification and termination.
✓ReadNorthgate_MA_v6.docx · 96 pages
✓ReadNorthgate_MA_v7.docx · 98 pages
●redline-helpercompare_documents(v6, v7)
Blocked by OpenAIDR78 ms
Northgate's outside counsel guidelines (§6.3) forbid sending its documents to outside tools without consent. redline-helper is not approved.
Understood. I'll compare the drafts here instead.
14 changes found. Indemnification: cap raised from 10% to 15% of the purchase price (§9.2). Termination: outside date moved to March 31, 2027 (§11.1), and a new reverse break fee (§11.3).
Reply to Claude…
OpenAIDR assistant
Ask about anything in this dashboard.What needs my attention?Summarize this finding
Where else is redline-helper installed, and has it ever received client files?
redline-helper is on 8 devices: 7 laptops and 1 document review server, docrev-02.
Only one call ever carried client files: today's call from alex-mbp, and it was blocked.
AI Supply ChainSession 4c2e-77b1Guardrail History
Draft a guardrail that blocks unapproved MCP servers on document review servers.
Proposed Guardrail change
scope: docrev-01 to docrev-04
+ deny: tool calls to MCP servers
+ unless review = Approved
+ reason: Unapproved MCP server
Tested on 30 days of activity: 0 normal actions blocked
Waiting for review
Ask a question…
9:12●●● ▮
OpenAIDRLeadership · Acme LLP
High9:12 AM
Northgate files nearly left the firm
Alex Chen's Claude Cowork tried to send two merger drafts to an unapproved tool.
Blocked before anything was sent.
Investigated; security team approved a new guardrail.
1 decision waiting for you
Do we have to tell Northgate?
No Northgate file left the firm: the only call was blocked before anything was sent. Alex's work is done, and the full record is ready if Northgate asks.
Session record
Decision for you
Require approval for new AI tools, firm-wide?
DeclineApprove
Recommendation accepted · sent to the security team
Thursday, October 29
9:12
TL;DR

OpenAIDR lets a law firm use AI agents safely, from the first install to a decision by leadership.

InstallOne app, every machineLaptops, the AI gateway and servers
InventoryEvery AI tool, sortedApproved, unapproved and high risk
PostureMisconfigurations foundEach one with its fix
GuardrailsRisky actions stoppedWritten from the firm's own rules
Detect and respondIncidents handledInvestigated, with an assistant
Leadership appLeaders informedMajor issues and a recommendation

Our example is Acme LLP, where lawyers use Claude Cowork and other AI agents every day.

InstallInstall onceLaptops, the gateway and servers
InventoryFinds every AI toolSorted and risk-rated by its models
PostureFixes weak setupsEvery misconfiguration, with its fix
GuardrailsStops risky actionsBlocks the action, not the work
Detect and respondHandles incidentsInvestigated on its own, with an assistant
Leadership appKeeps leaders informedA recommendation, then one tap
OpenAIDRa product of OpenSecura
OPENAIDR · LAW FIRM

What OpenAIDR does for a law firm, from the first install to a leadership decision

A self-playing demo, about two and a half minutes long. At Acme LLP, OpenAIDR is installed, finds every AI tool, fixes weak agent setups, stops risky actions, handles an incident with the security team, and brings the CISO a recommendation in the OpenAIDR leadership app.

Space plays or pauses; arrow keys step through sentences; the speaker button turns narration off.