When an AI tool prices a quote, flags a claim, or suggests a change to your records, Still Horizon captures what it saw, what it recommended, and who approved it — then seals that record so your auditor, insurer, or customer can verify it years later, without having to trust you or us.
Decision record
ACME-Q-2026-118204
Verified independently
e3b9 41c0 7a2f … 9d18 c6e4
Most AI governance tools watch who uses which AI tool and block what they shouldn't. That matters. But when a customer disputes a price, a regulator asks why an order shipped, or an insurer asks how you govern AI, a dashboard of policies doesn't answer the question. A record of the decision does.
| Typical AI governance tools | Still Horizon adds |
|---|---|
| See which AI tools people use | A record of each decision an AI tool influenced |
| Block or redact sensitive data in prompts | Proof of which data the decision used, and which version |
| Logs kept in the vendor's console | Sealed records anyone you authorize can verify on their own |
| Policies written for the AI tool | A named person's approval on every decision that needs one |
| A cloud service you connect to | Runs on your network, including fully offline sites |
Still Horizon works alongside the AI security tools you already have. It doesn't replace them.
Every AI decision moves through the same six steps. Choose a step to see what happens and what gets recorded.
Recorded
These are the failures that turn AI from an advantage into a liability, and what Still Horizon puts in their place.
Disputes, denials, and write-offs you can't defend.
Every decision recorded with the data and rule behind it.
Agents change records, prices, or limits with no approval trail.
AI proposes; a named person approves; both are sealed.
Errors repeat at machine speed until someone pulls the plug on everything.
Safe Stop halts exactly what's wrong and records it.
Logs that can be edited, deleted, or are stuck in a vendor's console.
Sealed records anyone you authorize can verify on their own.
Governance tools become one more place sensitive data can leak.
Runs on your network, including fully offline sites.
Exclusions and higher premiums when carriers can't see your controls.
Evidence packages underwriters can check.
What many teams call a kill switch, built for operations that can't simply go dark.
Why it matters now: organizations building their own models and agents need a tested, pre-authorized way to intervene, and regulators expect one. The EU AI Act requires that people overseeing high-risk AI be able to interrupt it.
A demo using ACME Hardware, a fictional company. Nothing here affects a real system.
Safe Stop record
StoppedSealed
ACME Hardware is a fictional glass and hardware distributor and fabricator: 200 employees, a 30-year-old ERP, an online store, a fabrication shop, and a sales team that started using AI for quoting last year. Here's what governed AI looks like there.
Open ACME's governance dashboard to try it yourself.
Seven months after the sale
A contractor disputes the price on a custom shower enclosure the AI quote assistant priced. Sales pulls the record: price list revision 4, inventory at that moment, the contractor discount rule, and the rep who approved it.
Settled the same afternoon, with the record attached to the reply.
Before an order ships
An AI order check flags that a door panel must be tempered safety glass and the order says annealed. The shop corrects it. The sealed record shows the check ran, what it found, and who fixed it.
If the job is ever questioned, ACME can show the safety check happened, and when.
Every day
An AI assistant proposes raising a customer's credit limit. It can't write to the ERP. The proposal goes to the credit manager, who approves it with a note, and the approval is sealed with the decision.
AI suggests. People decide. The record shows both.
Annual renewal
ACME's carrier asks how AI is governed before renewing the liability policy. ACME exports a year of AI pricing and order decisions: which rules applied, which were overridden, and who approved exceptions.
Evidence the underwriter can check, instead of a questionnaire answered from memory.
From day one
Cost, margin, and customer pricing rules are marked restricted. Each AI tool sees only what that person's role allows, and the records stay on ACME's own servers.
Pricing logic never leaves the building.
Every industry asks the same two questions about AI: how do we lower our liability, and how do we grow revenue with it? Choose yours, then switch between the value, the compliance picture, and the risks of running AI without governance. Every organization shown is fictional.
Manufacturing and distribution
What the rules ask for, and the evidence Still Horizon produces for it.
| Requirement | What it asks for | Evidence from Still Horizon |
|---|---|---|
| ISO 9001 quality records | Documented evidence that controls and checks were performed | A sealed record of every AI-assisted check and who resolved it |
| Product liability | Proof of reasonable care before a product shipped | Time-stamped evidence that safety and spec checks ran |
| Customer and insurer AI requirements | Answers to how AI is governed in your operations | Evidence packages instead of questionnaires answered from memory |
| CMMC (defense suppliers) | Controlled handling of covered defense information | Records stay on your network; access follows each role |
Still Horizon produces evidence for these requirements. It does not certify compliance, and this is not legal advice.
What tends to go wrong when AI runs in this industry without governance.
A disputed AI-generated price turns into a write-off because no one can show how it was calculated.
An AI check misses, or no one can prove it ran, and the liability claim lands on you.
An assistant changes credit limits or item data with no approval trail.
AI now prices quotes, checks orders, and suggests changes to the ERP. When a price is disputed or a product fails, you need to show what the AI was told, which rule it applied, and who signed off.
Decision record
Verified independently
Defense and government
What the rules ask for, and the evidence Still Horizon produces for it.
| Requirement | What it asks for | Evidence from Still Horizon |
|---|---|---|
| Federal AI governance requirements (OMB) | AI use-case inventories and minimum practices for high-impact AI, including human oversight | Decision records and approvals for each AI use, ready for inventory and review |
| NIST AI Risk Management Framework | Mechanisms to supersede, disengage, or deactivate AI that behaves outside its intended use | Safe Stop, with a sealed record of every stop and restart |
| Public records and retention laws | Records of official actions, retained and producible | Sealed records with retention you control |
| FedRAMP and CMMC environments | Data handled inside authorized boundaries | Deploys inside your boundary, including offline |
Still Horizon produces evidence for these requirements. It does not certify compliance, and this is not legal advice.
What tends to go wrong when AI runs in this industry without governance.
An inspector general asks who reviewed an AI recommendation, and the answer is a policy document.
A model drifts or misbehaves and there is no tested, authorized way to halt it.
Requesters ask how AI shaped a decision, and the records do not exist.
Agencies must inventory their AI uses, keep people in charge of consequential decisions, and answer to inspectors general and the public. A policy document is not evidence that oversight happened.
Decision record
Verified independently
Insurance and reinsurance
What the rules ask for, and the evidence Still Horizon produces for it.
| Requirement | What it asks for | Evidence from Still Horizon |
|---|---|---|
| NAIC model bulletin on AI use by insurers | A written AI program with governance, documentation and oversight of third-party AI | Records that show the program operating, decision by decision |
| Unfair claims practices laws | Fair, documented handling of each claim | Proof that an adjuster decided each AI-flagged claim, and on what grounds |
| State rules on algorithms and external data | Testing and evidence that models do not discriminate unfairly | Inputs and outcomes preserved for testing and review |
| Market conduct exams | Files that show how decisions were made | Evidence packages examiners can verify on their own |
Still Horizon produces evidence for these requirements. It does not certify compliance, and this is not legal advice.
What tends to go wrong when AI runs in this industry without governance.
A regulator or court finds no human review behind AI-driven claim decisions.
Policyholders use AI you cannot evaluate, so the only tool left is an exclusion.
You cannot reproduce which data drove an underwriting decision.
Insurers use AI in claims and underwriting, and are asked to underwrite their customers’ AI. Regulators expect governance of the first; carriers need evidence for the second instead of exclusions.
Decision record
Verified independently
Banking and financial services
What the rules ask for, and the evidence Still Horizon produces for it.
| Requirement | What it asks for | Evidence from Still Horizon |
|---|---|---|
| Model risk management guidance (SR 11-7) | Validation, monitoring and documentation of models in use | Ongoing records of each model-assisted decision and its outcome |
| Adverse action notices (ECOA / Regulation B) | Specific, accurate reasons for a credit denial, even when complex models are used | The exact inputs and reasons behind each decision, preserved |
| Third-party risk management | Oversight of vendor and AI services | Vendor-neutral records that work across every AI provider |
| Fair lending examinations | Evidence that decisions are consistent and explainable | Records examiners can test and verify |
Still Horizon produces evidence for these requirements. It does not certify compliance, and this is not legal advice.
What tends to go wrong when AI runs in this industry without governance.
An adverse action notice cannot be backed up by what the model actually used.
Examiners find AI in use with no monitoring or decision records.
A member asks why, and no one can answer.
Lenders must explain adverse decisions to applicants and show examiners that models are governed. When AI helps decide credit, the reasons and the data behind them have to be reproducible.
Decision record
Verified independently
Healthcare and life sciences
What the rules ask for, and the evidence Still Horizon produces for it.
| Requirement | What it asks for | Evidence from Still Horizon |
|---|---|---|
| HIPAA Security Rule audit controls | Mechanisms that record activity in systems with patient data | Sealed records kept on your network; patient data never leaves |
| Decision support transparency (ONC HTI-1) | Information about how predictive decision support works and is used | A record of what the AI suggested and which guideline version applied |
| State rules on AI in utilization review | Clinicians, not algorithms alone, make medical necessity decisions | Proof that a clinician made each call |
| FDA guidance for AI-enabled devices | Documentation of how AI functions perform and change | Records of suggestions and outcomes over time |
Still Horizon produces evidence for these requirements. It does not certify compliance, and this is not legal advice.
What tends to go wrong when AI runs in this industry without governance.
A prior authorization or triage decision traces back to a model alone.
A governance tool becomes a new place patient data can leak.
You cannot show what the AI suggested and what the clinician decided.
AI now suggests triage levels, drafts notes, and supports prior authorization. Clinicians stay accountable, so the record has to show what the AI suggested and what the clinician decided, without moving patient data off site.
Decision record
Verified independently
Legal and compliance
What the rules ask for, and the evidence Still Horizon produces for it.
| Requirement | What it asks for | Evidence from Still Horizon |
|---|---|---|
| ABA Formal Opinion 512 on generative AI | Competence, confidentiality and supervision when lawyers use AI | A record of AI use and the attorney who reviewed it |
| Discovery obligations | A defensible, explainable review process | Which documents AI reviewed, what it concluded, and who confirmed it |
| Court AI disclosure orders | Disclosure and certification of AI use in filings | Records that back up what you certify |
Still Horizon produces evidence for these requirements. It does not certify compliance, and this is not legal advice.
What tends to go wrong when AI runs in this industry without governance.
Opposing counsel challenges AI-assisted review and there is no record of how it was done.
AI output reaches a filing without documented attorney review.
Privileged material passes through tools you do not control.
When AI reviews documents or drafts filings, opposing counsel and courts can ask how it was used. Defensibility depends on showing the process, not describing it after the fact.
Decision record
Verified independently
The evidence Still Horizon produces lines up with what the major AI frameworks and regulators ask for. Industry-specific rules are in each industry's Compliance tab above.
| What Still Horizon produces | NIST AI RMF | EU AI Act | ISO/IEC 42001 | Sector regulators | Insurers and courts |
|---|---|---|---|---|---|
| Decision records | Documented, monitored AI outcomes | Record-keeping for high-risk systems (Art. 12) | Documented information on AI operation | Model risk, claims, clinical, and quality records | Proof of what was decided, and on what basis |
| Human approvals | Defined roles and accountability | Human oversight (Art. 14) | Roles, responsibilities, and authorities | Adverse action, claims, and clinical review | Proof a person made the call |
| Safe Stop records | Mechanisms to supersede, disengage, or deactivate AI (MANAGE 2.4) | Ability to interrupt the system (Art. 14) | Operational control of AI systems | Incident response and change control | Proof you could and did intervene |
| Sealed, verifiable evidence | Transparency and accountability | Logs kept and available to authorities | Internal audit and management review | Examinations and audits | Evidence that holds up in a dispute |
| Runs on your network | Secure, resilient, privacy-enhanced AI | Data protection alongside GDPR | Information security controls | HIPAA, CMMC, and FedRAMP boundaries | A smaller breach surface |
Still Horizon produces evidence that supports these requirements. It does not certify compliance, and this is not legal advice. Talk to us about the frameworks that apply to you.
Technical documentation and platform access are available to qualified organizations after a short review. Tell us what you need to govern and we'll reply within two business days.
Already a client? Sign in
Send the email that just opened and we'll reply within two business days. If no email opened, write to info@stillhorizon.ai.