Clinical safety infrastructure for the age of AI.
An independent check on every clinical order before it reaches the patient, whether a clinician wrote it or an AI tool drafted it.
The problem
Patient care is entering the age of AI, and nothing independent is checking its work.
AI tools already draft prescriptions, tests, and referrals from what they hear in the exam room, and their role in patient care is expanding fast. Every high-stakes industry that automated learned the same lesson: the system that acts and the system that checks it have to be separate.
Medicine has never had that second system. The safety checks built into electronic records work from written rules: known drug interactions, allergies, dose limits, duplicate orders. Checking an order against the whole patient, including the notes, the care plan, and care at other hospitals, is a different job. A clinician reviewing a stack of AI drafts at the end of the day is the last line of defense, and it is getting thinner.
Detection
The errors that get through today.
Four examples. Two orders were drafted by AI tools and two were written by clinicians. In every case, the information needed to catch the problem was already in the patient's record.
AI-drafted · Held
A discussion became an order.
During a visit, the physician and patient talk about possibly stopping a blood pressure medicine. The AI tool drafts an order to stop it, and the order is signed with the rest of the day's drafts. Nothing in the record supports the change. The patient's blood pressure, controlled for years, would have gone uncontrolled.
At signing, the order is checked against the chart and the notes. There is no basis for stopping the drug. The order is held and sent to the clinician the hospital's policy names.
AI-drafted · Held
A fact the record contradicts.
The physician says the patient's kidneys have been fine, and the AI tool drafts a full dose of a drug the kidneys have to clear. The local lab results are more than a year old. A kidney test from nine days ago, done at another hospital, shows they are not fine. The result existed. It was in a different building.
At signing, the dose is checked against the newest result anywhere in the record. The order is held for dose review before it reaches the patient.
Human · Held
Two good orders that cancel each other.
A cardiologist orders a drug to pull fluid off a patient (IV furosemide). Two hours later, a kidney specialist, unaware, orders a saline drip to add fluid back for a rising kidney test. No rule covers this pair. Each order makes sense on its own. Together they work against each other.
The conflict is identified before the second order takes effect. It is held and routed for the two teams to coordinate, in the way the hospital's policy defines.
Human · Cleared
An order that looks wrong and is right.
A patient on methotrexate is prescribed folic acid. On paper the two work against each other, and many rule-based systems raise an alert every time the prescription is refilled.
Because the third check reads the whole chart, it recognizes a standard, intentional combination. The order passes silently. Nobody is interrupted.
Product
OrderGuard checks every order against the whole patient before it takes effect.
OrderGuard is an independent safety check for clinical orders. Every order that would change a patient's care, whether a clinician wrote it or an AI tool drafted it, is checked against the patient's complete record in under a second. Orders that are safe under your policy pass silently. The rare order that meets your threshold is held at the point of signing until a clinician resolves it.
Quiet by design.
Designed for about five holds per thousand orders. A clinician should see a hold a couple of times a month.
Nothing to dismiss.
A hold stays until the person your policy names resolves it, in the screen they already use.
Your policy.
The hospital decides which findings hold an order, who resolves them, and how. OrderGuard applies the policy. Clinicians make the decisions.
How an order is checked
Facts
A direct comparison with the record. Does the evidence exist? Is the claim true? Was a fact hallucinated? Is there a newer result somewhere else in the patient's record?
Collisions
A mathematical model of the patient's care. Do two things in this patient's care work against each other, even when no rule names the pair?
Context
Reasoning, used only on orders the first two checks have flagged. Is the conflict a real risk, or a documented, deliberate decision? Most orders never reach this step.
Facts, then conflicts, then judgment.
First, the facts: does the record support what the order says? Second, the conflicts: does the order work against anything else in the patient's care? Third, only for orders those two flag, a reading of the whole chart to judge whether the conflict is real. The first two checks are math. Only the third uses generative AI.
What OrderGuard sees.
The order as it is signed, and the patient's whole record, including care at other hospitals. Most drafting mistakes show up right there: a diagnosis the chart never mentions, a drug the patient already takes, a dose the last kidney test rules out. The rare mistake that leaves no trace in the order is covered on the Standard page.
When an order is held.
The hold goes to whoever the hospital's policy names for that kind of finding: the ordering clinician, the attending, a pharmacist, or a committee. They correct the order or document why it stands. Every hold and every resolution is written to a permanent record.
The Standard
What an independent check has to be.
Independent clinical order verification means that every order that would change a patient's care is checked before it takes effect. The check compares the order with the patient's record at the hospital and with the patient's history from other organizations, retrieved through the national health information exchange (TEFCA). It is performed by a system independent of the clinician who wrote the order, the developer of any software that drafted it, the platform that carries it, and the insurer who pays for the care. Orders are cleared or held on written criteria, and every check, hold and resolution is recorded.
Every high-stakes industry settled in the same place for the same reason. Banking, financial markets, and aviation all keep the system that acts separate from the system that checks it. Medicine is the last one without that separation.
A check works for every party when it is owned by none of them.
The record
Every order leaves a record.
Every order OrderGuard checks produces a time-stamped entry in the Safety Compliance Ledger: who or what wrote the order, what was checked, what was found, whether the order was cleared or held, who resolved it, and how. Entries are never edited or removed.
Over time the hospital holds an unbroken record of due diligence, order by order, that its board, its lawyers, its insurer, and its accreditor can rely on.
Illustrative. Order identifiers are synthetic. Source is recorded when the drafting software marks its orders.
Alignment
Built for every party that carries the risk.
Health systems
Clinical authority stays with your clinicians. Every hold is resolved inside your own process, at the level your policy sets. Your board and general counsel get independent verification they can see, paid for as an operating expense. It is also what lets a board say yes to AI-drafted ordering.
Health plans
Every prevented adverse event improves quality scores and shared savings. A neutral check a plan can rely on precisely because the plan does not run it.
Malpractice carriers
A documented, order-level record of due diligence across every insured facility. The precedent is the surgical checklist: an independent safety practice a carrier can recognize in underwriting and rely on in defending a claim.
Accreditors and regulators
An auditable record of every order's check, hold, override, and resolution, kept by a party independent of the hospital, the record system, and the payer. Current guidance already calls for independent monitoring and safety-event reporting. This is that record.
Deployment
Live in days, on the standards hospitals already use.
OrderGuard connects through the interfaces built into every major electronic record system (FHIR and CDS Hooks) and reaches records from other hospitals through the national exchange network (TEFCA). There is no EHR build. A hospital's own IT team configures it, and technical integration is measured in days. Designing the governance policy is the real work.
Every hospital on the network uses the same verification model. Each hospital's patient records stay under that hospital's governance and agreements, which also define how outcomes strengthen the shared model. What the network learns about a kind of error protects every hospital on it.
Your policy
- You decide which orders are checked, the thresholds at which an order is held, and who resolves each kind of hold.
- Holds are rare by design.
- OrderGuard runs alongside your existing safety tools.
Integration
- Works with Epic, Oracle Health, and Meditech.
- Standard interfaces only. No EHR build.
- Configured by your own IT team.
Economics
- Priced per order checked.
- No license fee and no implementation cost.
- An operating expense that scales with the orders it protects.
Origin
Cirtiq was founded after co-founder Chris Riley's thirteen-year-old son suffered a massive stroke. In the years that followed, across three states, the family watched orders written that contradicted his own medical history, buried deep in the record. No one was negligent. The system had no independent check between a decision and its execution. Cirtiq exists to be that check.
Every clinical order, verified.
Start a conversationRead the Standard · Press: press@cirtiq.com