Find the AI you run, rate the risk, and map controls only to the rules that apply to you.
The first governance question is simpler than it sounds. Which AI systems do we run, and who owns each one? Many teams can't answer it on the first try, because AI arrives inside SaaS products as often as it is built. We build the inventory first, then the controls.
Governance comes down to knowing which AI systems you run, who owns them, how their risk is assessed and which rules apply.
We build the inventory first. Then come the risk assessments and controls mapped to the NIST AI RMF, ISO/IEC 42001 and the EU AI Act, while we track US state laws such as Colorado's. We also test the systems for prompt injection, data leakage and over-permissioned agents.
you probably need this if
Ask for a list of the AI in use, with owners and the data each one touches, and nobody can produce one.
There's a customer questionnaire asking about NIST AI RMF or ISO/IEC 42001, and nobody's got an answer ready.
You sell in the EU or Colorado, and nobody can say for sure which AI rules your products fall under.
terms on this pagePrompt injectiondefinitionInstructions tucked into content an AI reads, like a document or a ticket comment, meant to make it do something its user never asked for.explained in a field note →
From the first call to a setup your own team can run.
Here's what usually prompts a call about your AI Governance & Security, what we do about it and in which order, and where you end up.
01 · where it hurts02 · what we change03 · how it runs04 · where you end up
incident board · ai governance & security3 open
openAsk for a list of the AI in use, with owners and the data each one touches, and nobody can produce one.
recurringThere's a customer questionnaire asking about NIST AI RMF or ISO/IEC 42001, and nobody's got an answer ready.
blockedYou sell in the EU or Colorado, and nobody can say for sure which AI rules your products fall under.
exposure
AI system inventory and ownershipNIST AI RMF and the Generative…ISO/IEC 42001 readinessEU AI Act readinessUS state AI lawsAI security testing
startrehearsehandover
Inventory
Classify
Design controls
Operate
we are here
ready to hand over
There's a named owner and a documented purpose now.
The next questionnaire gets a real answer, a pointer to a running management system and…
We had the mark on the output before 2 December, and the tests behind the choice of…
AI system register with owners and risk ratings
chapter 01 · where it hurts
Why teams reach out
Ask for a list of the AI in use, with owners and the data each one touches, and nobody can produce one.
There's a customer questionnaire asking about NIST AI RMF or ISO/IEC 42001, and nobody's got an answer ready.
You sell in the EU or Colorado, and nobody can say for sure which AI rules your products fall under.
chapter 02 · what we change
How we solve it
AI system inventory and ownership
NIST AI RMF and the Generative AI Profile
ISO/IEC 42001 readiness
EU AI Act readiness
US state AI laws
AI security testing
chapter 03 · how it runs
Our approach
Inventory
Classify
Design controls
Operate
chapter 04 · where you end up
What changes
There's a named owner and a documented purpose now.
The next questionnaire gets a real answer, a pointer to a running management system and its records.
We had the mark on the output before 2 December, and the tests behind the choice of method are on file for anyone who asks.
AI system register with owners and risk ratings
what we do
Where we spend the hours
6 areas of work, each one done by an engineer who has run it in production.
We list the AI you built and the AI you bought, each with owner, purpose, data, model provider and the decisions it touches. SaaS features count too. That's where most of the unreviewed use is.
We organized the controls around the RMF's Govern, Map, Measure and Manage functions. AI RMF 1.0 came out in January 2023, the Generative AI Profile (NIST AI 600-1) in July 2024, and NIST says the RMF is being revised now. We're tracking that.
We build an AI management system to ISO/IEC 42001:2023. It covers policy, roles, risk and impact assessment and the Annex A controls, in good enough shape for an accredited certification body to audit.
Some of the AI Act already applies. Prohibited practices since February 2025, general-purpose AI model duties since August 2025. The high-risk dates moved. The Digital Omnibus, Regulation (EU) 2026/1744, pushed Annex III obligations to 2 December 2027 and Annex I to 2 August 2028.
SB 24-205 never took effect. SB 26-189, signed in Colorado on 14 May 2026, repealed it and reenacted it as an automated decision-making technology law. Its core is notice, explanation of adverse outcomes, and human review.
Testing against the OWASP Top 10 for LLM Applications 2025, prompt injection, sensitive information disclosure and excessive agency among them. We also check what each agent's credentials can actually reach.
capability 01 / 06in production
AI system inventory and ownership
We list the AI you built and the AI you bought, each with owner, purpose, data, model provider and the decisions it touches. SaaS features count too. That's where most of the unreviewed use is.
We organized the controls around the RMF's Govern, Map, Measure and Manage functions. AI RMF 1.0 came out in January 2023, the Generative AI Profile (NIST AI 600-1) in July 2024, and NIST says the RMF is being revised now. We're tracking that.
We build an AI management system to ISO/IEC 42001:2023. It covers policy, roles, risk and impact assessment and the Annex A controls, in good enough shape for an accredited certification body to audit.
Some of the AI Act already applies. Prohibited practices since February 2025, general-purpose AI model duties since August 2025. The high-risk dates moved. The Digital Omnibus, Regulation (EU) 2026/1744, pushed Annex III obligations to 2 December 2027 and Annex I to 2 August 2028.
SB 24-205 never took effect. SB 26-189, signed in Colorado on 14 May 2026, repealed it and reenacted it as an automated decision-making technology law. Its core is notice, explanation of adverse outcomes, and human review.
Testing against the OWASP Top 10 for LLM Applications 2025, prompt injection, sensitive information disclosure and excessive agency among them. We also check what each agent's credentials can actually reach.
We pieced these together from AI Governance & Security situations that come up a lot. They're patterns. You won't find a specific client in them.
case file 01
Applicant-ranking tool used for Colorado hires
composite · not a client story
Beforethe situation
Recruiters only see applicants after a vendor tool has ranked them, and some of the roles are in Colorado. Nobody has written down how the ranking works or what data feeds it. If an applicant is rejected, there's no way to ask a person to look again.
What we dothe work
The tool went into the AI register first. With your counsel we mapped it against SB 26-189 and asked the vendor for its documentation. We drafted the applicant notice and the adverse-outcome explanation, designed the human review path, and tested outcomes across applicant groups where the historical data allowed.
Afterwhere you end up
There's a named owner and a documented purpose now. Counsel decides compliance from that file, and it's reusable for EU AI Act work if hiring ever reaches the EU.
case file 02
ISO/IEC 42001 questions showing up in security questionnaires
composite · not a client story
Beforethe situation
Enterprise buyers have started asking in security questionnaires whether the product's AI features follow ISO/IEC 42001. LLM features ship monthly. There is an ISO 27001 ISMS already, but no AI policy of any kind.
What we dothe work
Rather than stand up a second management system, we extend the ISMS you have: AI scope and policy, an impact assessment process, and whichever Annex A controls apply. Each release picks up a model approval step with an evaluation record. Internal audit runs before you talk to a certification body.
Afterwhere you end up
The next questionnaire gets a real answer, a pointer to a running management system and its records. Whether to certify is now a choice about a gap you already know about.
case file 03
EU image generation and the 2 December 2026 marking date
composite · not a client story
Beforethe situation
Marketing images and copy for EU customers have come out of this product since early 2026. That puts it on the market before 2 August 2026, so under the Digital Omnibus it has until 2 December 2026 to comply with the Article 50(2) marking duty.
What we dothe work
Once counsel confirmed scope, we picked a machine-readable mark for each output format it could carry, embedded metadata or a watermark for example, and checked whether it survived typical edits and exports. The method and test results went into the technical documentation.
Afterwhere you end up
We had the mark on the output before 2 December, and the tests behind the choice of method are on file for anyone who asks.
How an AI governance program fits together from inventory to evidence
We put every AI system in the inventory with an owner, then assess its risk under the NIST AI RMF and the laws that apply. Controls like access scoping and security testing follow from the rating, and what they produce is the evidence auditors and customers ask for.
01 · Inventory
AI system inventoryowner · purpose · data used → Risk assessment · each system
Agents and identitiesEntra ID · Okta · SailPoint → AI system inventory · non-human identities
02 · Risk method
Risk assessmentNIST AI RMF · GenAI Profile → Access scoping · control set→ AI security testing · control set→ Monitoring · control set
Applicable rulesEU AI Act · Colorado AI Act → Risk assessment · obligations
AI system inventory → Risk assessment: each system
Agents and identities → AI system inventory: non-human identities
Applicable rules → Risk assessment: obligations
Risk assessment → Access scoping: control set
Risk assessment → AI security testing: control set
Risk assessment → Monitoring: control set
Access scoping → Evidence pack: review records
AI security testing → Evidence pack: findings
Monitoring → Evidence pack: logs
how it runs
A typical engagement
We don't start a step until the last one has produced something you can check.
step 01
Inventory
Find the AI systems. Procurement records, SSO and OAuth app grants, network logs and interviews all turn some up. Then assign owners.
step 02
Classify
A risk rating per system, and a note of which legal regimes apply. The legal reading is your counsel's to confirm.
step 03
Design controls
We size policies, approval workflow, testing requirements and monitoring to each risk tier. Low-risk tools shouldn't be buried in paperwork.
step 04
Operate
After launch it's ongoing work. Periodic reviews, incident handling, evidence collection, and a tracker that follows regulatory changes as they happen.
you keep
What you're left with
We write these for your AI Governance & Security setup, and they're working documents. Your team should be able to keep things running from them long after we've gone.
inventory01
AI system register with owners and risk ratings
yours to keep
report02
Regulatory applicability analysis per system
yours to keep
evidence03
Control set mapped to NIST AI RMF functions and ISO/IEC 42001 Annex A
Some clients want a project team, some want a couple of engineers on theirs. Either works. Onsite, nearshore or offshore, with handovers across time zones that keep things moving.
Onsite
Engineers at your sites in the US
Nearshore
Overlapping hours with US teams
Offshore
Overnight progress on your backlog
All time zones
Coverage handed over across time zones
24-hour coverage · US Central time
OnsiteUS business hours
Nearshoreoverlapping US hours
Offshoreovernight progress
Combinedhandover, not hand-off
00:0006:0012:0018:0000:00
roles we staff
AI governance lead
AI security engineer
Model risk analyst
Identity engineer for AI agents
Data protection engineer (DLP, classification)
due diligence · ai governance & security
Questions to ask any partner, including us.
Put these to anyone bidding on your AI Governance & Security work. The ones who've done it answer differently from the ones who've read about it.
Q1
How will you find the AI features inside SaaS products we already pay for?
Much AI arrives through vendor updates rather than projects. An inventory that only covers internal builds misses the systems most likely to surprise you.
Q2
Do your controls produce evidence on their own, or does someone have to remember?
What auditors and certification bodies look for is records. Put the approval step in the release pipeline and the trail writes itself. A wiki policy leaves nothing behind.
Q3
How are you going to track changes like the Digital Omnibus, or Colorado's rewrite of its AI law?
Dates and obligations moved several times in 2026. A governance program needs a named owner and a source list for regulatory changes.
Q4
How do you test an agent for excessive agency, beyond prompt injection?
Excessive agency has its own entry in the OWASP list. Look at the tools and permissions an agent holds. What it can be tricked into saying is only part of it.
Q5
Where does your work stop and our counsel's begin?
Applicability is a call for lawyers. If a partner blurs that line, you carry the risk. Map the systems and controls, and let counsel decide with that in hand.
01When do the EU AI Act's high-risk rules apply now?
Later than planned for high-risk systems. The Digital Omnibus (Regulation (EU) 2026/1744) has been in force since 27 July 2026. Annex III high-risk systems, like those used in employment or credit, now have obligations from 2 December 2027. Systems covered by Annex I product legislation follow on 2 August 2028. Article 50 transparency duties weren't postponed, though generative systems on the market before 2 August 2026 have until 2 December 2026 for the Article 50(2) marking duty.
02What happened to the Colorado AI Act?
Colorado changed it twice. SB 24-205's start date moved to 30 June 2026, and then SB 26-189, signed on 14 May 2026, repealed and reenacted the law. What's left applies to automated decision-making technology used in consequential decisions. You owe consumers notice, a plain-language explanation within 30 days of an adverse outcome and a path to human review. Developer documentation duties start 1 January 2027. The attorney general enforces it, with no private right of action.
03Is the NIST AI RMF mandatory?
No, it's voluntary. NIST publishes it for anyone to use. We'd adopt it anyway, because customers, regulators and auditors increasingly ask how you manage AI risk and the RMF gives you shared terms to answer in.
04Do we need ISO/IEC 42001 certification?
Only if customers or contracts ask for it. Plenty of organizations build the management system to the standard first and decide about certification later. An accredited certification body does the audit, and we get you ready for it.
05Do you give legal advice?
No. We map systems and controls to the regulations and work alongside your counsel, who makes the legal calls.
related
Often paired with
Projects on AI Governance & Security usually end up touching one of these, so they're worth a look.
