Assay integrations
Fits into the systems you already run.
Assay keeps track of the AI your agency uses, who signed off on each tool, and what it can get into. It’s meant to sit alongside the tools your teams already work in, not replace them.
Today it signs people in with the accounts they already have, finds AI apps across your Microsoft environment, and hands its audit trail to your SIEM by API, webhook or syslog. Next come ready-made Sentinel and Splunk packages and your ticket systems.
Your sources
- Microsoft Entra ID · sign-in, app discovery
- Other OIDC providers · sign-in
- Defender for Cloud Apps · AI services in traffic
- Spreadsheet inventory · CSV import
- Microsoft Purview · AI activity
- Your own AI apps · OpenTelemetry
Assay
Adapters, the register and a hash-chained audit trail, served through one public API and event feed.
Your tools
- Microsoft Sentinel · Content Hub solution
- Splunk · add-on or HEC
- QRadar, syslog SIEMs · TLS and CEF
- ServiceNow, Jira · a ticket per finding
- Archer, OneTrust · registry feeds
- Firewalls, proxies · block-list feed
- Admin scripts · PowerShell, Python
- Residents · public AI register
What your tools can’t tell you
Your agency already runs tools that find AI, protect data and keep records. Each one answers a technical question. None of them records whether a particular use of AI was allowed, why, and on what terms.
That’s the part Assay adds. It reads what your tools find, records what your agency decides, and hands those decisions back for them to enforce.
Identity and sign-in
Who is this person, and what can they reach?
What Assay adds
Who owns each AI use, and who may approve it.
AI discovery
Which AI tools are people using?
What Assay adds
Whether each one is approved, restricted or blocked, and for what.
Data protection
Is sensitive data going into AI?
What Assay adds
Which data each approved use may handle, as a condition of approval.
Records retention
What did people type, and can we produce it?
What Assay adds
Why a use was approved: the review, sign-offs and evidence.
Compliance checks
Do our controls meet the framework overall?
What Assay adds
Which controls apply to each use at its risk level, with evidence.
Security monitoring
Is anything suspicious happening?
What Assay adds
A record of every governance change that your SIEM can check itself.
Firewalls and proxies
Can this traffic go through?
What Assay adds
The decision about what to block, published for them to enforce.
Tickets and registries
Who is fixing this, and is it logged?
What Assay adds
The finding, its deadline, and the evidence that it was fixed.
Your tools see what AI touches. Assay records whether each use is allowed, why, and what you’ve told the public.
How we build every connection
Governance software has a way of turning into one more system someone has to babysit, and integrations are usually where that starts. So we hold every connection we build to the same few rules.
It all runs through the public API
Our own connectors use the same API we document for you. If we can build something on it, so can your team, and nothing depends on a back door only we know about.
Nothing gets installed in your AI apps
We won't ask you to add an agent or a library to the apps you run. Microsoft data comes through Microsoft's own APIs, and apps you build can report through OpenTelemetry, which they may already use.
Your tools stay in charge of blocking
Assay publishes its decisions for your firewall or proxy to act on, and only reads from your Microsoft tenant. If Assay ever goes down, nothing gets blocked because of it.
An audit trail your SIEM can check for itself
Assay records every change people make, sign-ins and approvals included, and links each entry to the one before it. If anyone edits or deletes an entry, the chain breaks.
The event feed hands that trail to your SIEM in order. Your security team can spot a gap on its own, without having to take our word for it.
- Your SIEM can pull
- It asks for anything new since the last event it saw. We don't need any of your passwords, nothing has to reach into your network, and if it's offline for a week it catches up.
- Or we can push
- To a webhook or an encrypted syslog collector today, with Splunk's HTTP Event Collector and Azure's Logs Ingestion API next. If a delivery fails we keep retrying for a day, then pause and hold your events until you're ready.
- In a format it already knows
- Events follow the OCSF 1.3 standard. Our Sentinel and Splunk packages, coming next, translate them into each SIEM's own model: ASIM for Sentinel, CIM for Splunk.
{ "class_uid": 3002, "class_name": "Authentication", "activity_name": "Logon", "status": "Success", "time": 1792591653000, "message": "Signed in", "metadata": { "version": "1.3.0", "uid": "48212", "event_code": "auth.signed_in" }, "user": { "uid": "214", "name": "Rosa Alvarez" }, "unmapped": { "assay": { "action": "auth.signed_in", …, "prev_hash": "47b0e9a3…c3d9" }, "hash": "9f2c71d0…e41a" }}
What connects today, and what’s next
We’d rather be clear than impressive here. Below is what works now, what we’re building and in roughly what order, and what we’ll build as soon as an agency needs it.
Working today
These are in Assay now.
- Sign-inWorks today
Single sign-on
Staff sign in with the work accounts they already have, through Entra ID or any OpenID Connect provider. Their roles come from your directory groups.
Brings data in · OpenID Connect
- DiscoveryWorks today
Entra ID apps
Shows which AI apps people have connected to their work accounts, what those apps can reach, and who actually signs in to them.
Brings data in · Microsoft Graph
- DiscoveryWorks today
Defender for Cloud Apps
Pulls in the AI services Defender has spotted in your network traffic, including the ones nobody told you about.
Brings data in · Microsoft Graph
- DiscoveryWorks today
Spreadsheet import
Already tracking AI tools in a spreadsheet? Bring it in and you're not starting from zero.
Brings data in · CSV upload
- ReportingWorks today
Public AI register
Publishes the AI uses you've approved so residents can see them, as a web page and as data others can reuse.
Sends data out · Web page and JSON
- DevelopersWorks today
Public API
The same API our own connectors use, documented in an OpenAPI file so your team can build on it too. Each tool gets its own client and hour-long tokens.
Sends data out · REST, OpenAPI 3.1
- DevelopersWorks today
Event feed and webhooks
Everything Assay records, in order. Your tools can pull it, or we can push it to them with a signature, and either way they can check nothing was altered.
Sends data out · REST, Standard Webhooks
- SIEMWorks today
QRadar and syslog SIEMs
Sends events over encrypted syslog, as CEF or JSON, for on-premises and older SIEMs.
Sends data out · Syslog over TLS, CEF
- BlockingWorks today
Firewalls and proxies
Publishes the AI services you've decided against at a private URL your firewall or proxy checks on a schedule. A decision shows up on its next fetch.
Sends data out · Block-list feed
- BlockingWorks today
Email notices
A daily email to each person with what needs them: overdue findings, sign-offs waiting, reviews coming due. People can turn it off for themselves.
Sends data out · Email
- ReportingWorks today
OMB AI use-case inventory
Exports your inventory with OMB's exact 32 columns. Assay fills in what it holds and leaves the rest blank rather than guess.
Sends data out · CSV export
- DevelopersWorks today
PowerShell and Python
Ready-made clients for the scripts your admins already write, with nothing else to install. They're generated from the API, so they never fall out of step.
Sends data out · PowerShell module, Python package
Being built
Roughly in this order. Most are built on the public API, which already works.
- Sign-in
SAML sign-in
For directories that use SAML rather than OpenID Connect.
Brings data in · SAML 2.0
- Sign-in
User provisioning
New hires, transfers and departures sync from Entra ID or Okta on their own. When someone leaves, they're signed out right away.
Brings data in · SCIM 2.0
- SIEM
Microsoft Sentinel
A Sentinel solution that brings Assay's events into Sentinel's standard audit table, with a workbook and alert rules ready to go.
Sends data out · Codeless connector
- SIEM
Splunk
A Splunkbase add-on that maps Assay's events to Splunk's data models, or sends them straight to your HTTP Event Collector.
Sends data out · Add-on or HEC
- Blocking
Teams notices
The same reminders in Microsoft Teams, for agencies that live there.
Sends data out · Teams
- Tickets and GRC
ServiceNow
Opens a ticket for each finding and shows its status back in Assay. There's nothing to install on the ServiceNow side.
Sends data out · Table API
- Tickets and GRC
Jira
The same thing for Jira, Cloud or Data Center.
Sends data out · REST API
- Tickets and GRC
Archer
If your state registers AI in Archer, Assay keeps that entry current and holds the review and evidence behind it.
Sends data out · REST API
- Tickets and GRC
OneTrust
Keeps OneTrust's AI inventory in step with the decisions made in Assay.
Sends data out · Inventory API
- Reporting
State inventory formats
Matches each state's reporting format with a column map instead of custom code.
Sends data out · CSV export
- AI activity
Microsoft Purview
Shows how staff actually use Copilot and other Microsoft AI, from audit data you're probably already collecting.
Brings data in · Purview audit
- AI activity
Azure API Management
Tracks the calls your own applications make to AI models.
Brings data in · Log Analytics
- AI activity
Your own AI apps
Apps your team builds can report their AI use through OpenTelemetry. You won't need to add any code from us.
Brings data in · OpenTelemetry
When you need it
Ready to build as soon as an agency asks. If that's you, tell us and it moves up the list.
Okta, Google Workspace, web gateways
The same AI discovery for agencies that aren't on Microsoft.
Defender for Cloud Apps tags
Marks apps as approved or not in Defender. Assay only reads from your Microsoft tenant, so this would be a separate permission you choose to grant.
Google SecOps
Sends events to a SecOps feed, which reads them with its built-in OCSF parser.
Elastic
An Elastic integration that reads the event feed.
Amazon Security Lake
Writes events to your Security Lake bucket as a custom source.
OSCAL
Control evidence in the format federal compliance tools read.
Keys and network access
We keep as few of your keys as we can
All of this works today, except two things that depend on where Assay is hosted: a managed identity, and fixed outbound addresses.
- Certificates instead of passwords
- Assay connects to Microsoft with a certificate. If you run it in your own Azure Government subscription, it can use a managed identity and hold no secret at all.
- Your tools connect to us
- Your SIEM and firewall reach out to Assay the same way they reach any cloud service, so you don't have to open anything up.
- Tokens that expire in an hour
- Each tool gets its own API client with only the permissions it needs. Clients sign in with a key pair where they can, and their tokens last an hour.
- Every tool has a name
- API clients show up in the audit trail by name, like "Splunk add-on," so you can see exactly what each one changed.
- Signed messages
- Webhooks carry an HMAC-SHA256 signature, following the Standard Webhooks spec, so your receiver can confirm they really came from Assay.
- Predictable addresses
- Assay calls out from fixed IP addresses you can allow-list, and only to the systems you've set up.
We don’t call a connection done until it passes a real test
We decide what the test is before we start building, and we put it here so you can hold us to it. Nothing moves to “working today” until it passes.
Passed 24 Oct 2026
Event feed
We point a stand-in SIEM at Assay, let it read the whole audit trail, then keep it following along live. Partway through, we kill the delivery worker. Nothing can go missing or show up twice.
Passed 24 Oct 2026
Firewall block list
A decision made in Assay has to show up in the list your firewall pulls, on its very next fetch.
Passed 24 Oct 2026
OMB report
Our OMB export's columns have to match OMB's published field list, word for word.
Passed 24 Oct 2026
Scripting clients
The PowerShell and Python clients run their tests against our demo site every time we release.
Not passed yet
Sign-in and provisioning
Microsoft's and Okta's own SCIM tests have to pass. And when someone is deactivated in your directory, they're signed out of Assay right away.
Not passed yet
SIEM packages
Each one has to work in a real Sentinel workspace and a real Splunk instance, and pass the checks Microsoft and Splunk run before they'll list it.
Not passed yet
Tickets
A finding has to make the full trip through a ServiceNow test instance and back.
Tell us what you’re running
Send us your identity provider, your SIEM and your ticket system. We’ll tell you honestly what works with Assay today and what’s still being built.
Get in touch