What Your Malpractice Carrier Actually Says About AI-Generated Notes

The question physicians don't ask early enough

When physicians evaluate AI medical scribes, the questions they ask are usually about workflow, accuracy, and pricing. The question they don't ask until later, and sometimes not until after a claim, is "what does my malpractice carrier think about this?"

The answer matters because malpractice carriers have updated their stance on AI-generated notes over 2024 to 2026, and the stance varies meaningfully between carriers. Some have explicit guidance. Some are silent and will form a position at claim time. Some have informal preferences that aren't published but are applied during underwriting.

This article covers what the major US and Canadian carriers actually say in 2026, what to disclose to your carrier before adopting an AI scribe, and what to do if a claim involves an AI-generated note.

What carriers actually say about AI scribes in 2026

The major carriers have published positions ranging from "no specific guidance" to "explicit acceptance with conditions." A representative sample:

The Doctors Company (US). Issued guidance in late 2024 stating that AI scribes are acceptable when used as documentation aids, provided the physician reviews and signs every note, retains responsibility for accuracy, and discloses the use to patients per state law. No premium adjustment for using AI scribes either way.

CNA (US). Published a 2025 risk-management advisory on AI scribes. Recommends practices document the consent process, retain audit trails of who reviewed each note, and verify the AI vendor's data handling complies with HIPAA. Considers AI scribes a "neutral risk factor" if used per their guidance.

MedPro Group / Berkshire Hathaway (US). Has not published specific guidance as of early 2026. Underwriters have informally expressed comfort with AI scribes used with physician review, but the lack of published guidance means each claim is evaluated case-by-case.

ProAssurance (US). Published acceptance with conditions in 2025. Their guidance emphasizes the importance of provider review, contemporaneous note signing, and proper consent. They've also recommended practices retain the original audio recording for at least 60 days for claim defense purposes; this is a different recommendation than what most AI scribe vendors default to.

Canadian Medical Protective Association (CMPA). Has issued guidance treating AI scribes as similar to other clinical documentation tools, with emphasis on physician responsibility for the final note, informed patient consent, and ensuring the vendor meets provincial privacy law (PHIPA, PIPEDA, Quebec Law 25). Their guidance specifically notes that an AI-generated note that the physician signs becomes the physician's note for medico-legal purposes.

HIROC (Canadian hospitals). Has supported member hospital pilots of AI scribes. Their position is similar to CMPA's: physician is responsible for the note, vendor must meet privacy law, consent must be documented.

The common thread across carriers: AI scribes are acceptable when used responsibly. The physician's review-and-sign step is the load-bearing piece. The vendor relationship needs to meet privacy law. Patient consent needs to be informed.

What to disclose to your carrier before adopting

Most carriers don't require pre-approval to use an AI scribe, but proactive disclosure is good practice and is often required when a carrier asks about technology changes during renewal. What to disclose:

The vendor name and the type of tool. "We are using [Vendor] as an ambient AI scribe that records visit audio and generates draft clinical notes for physician review and signature."

The data handling. Where data is stored, how long, and the vendor's compliance posture (HIPAA, PHIPA, PIPEDA, depending on jurisdiction).

The consent process. How you obtain patient consent, what the script is, how you document the consent.

The physician review process. Confirm that every AI-generated note is reviewed and signed by the physician before being incorporated into the record.

The audit trail. Confirm what records you keep: who reviewed what, when, what edits were made, original audio retention (if any).

Send this as a one-page summary to your carrier's risk management contact. Most carriers will acknowledge receipt and tell you whether they have any specific concerns. The disclosure itself becomes a defensive document if a claim ever requires demonstrating you followed reasonable adoption practices.

What changes about your liability with an AI scribe

The honest answer: not much, but the failure modes shift.

What doesn't change:

You remain legally responsible for the contents of every note you sign. An AI-generated note that you sign is, for legal purposes, your note. You cannot defend a misdiagnosis by saying "the AI documented the case differently than I assessed it." The note is what you put your name on.

What does change:

The kinds of errors that show up in notes shift. Human-typed notes typically contain typos, abbreviated phrases, occasional missed sections. AI-generated notes typically contain occasional misheard medications, occasional conflation of conditional discussions with confirmed orders, and occasional drift from the provider's exact intent. The error rate isn't necessarily higher; the patterns are different.

This pattern shift affects how plaintiff attorneys approach a chart in litigation. Some carriers report that plaintiff attorneys have learned to look for specific AI-error patterns (the inconsistent dose, the over-formal phrasing, the suspiciously thorough HPI for a 10-minute visit). Defense in cases where the chart shows AI-pattern errors sometimes requires explaining the technology, which juries vary widely in their comfort with.

The mitigation is straightforward: the same physician review that prevents errors also prevents the litigation pattern.

What to do if a claim involves an AI-generated note

You discover (often during chart review months after the fact) that a note you signed contained an error from the AI scribe. The error is now part of the record. A patient has filed a complaint or a claim is pending.

Within the first 24 hours:

• Contact your carrier. Don't wait. Carriers want to be involved early; late notification can affect coverage.
• Preserve everything. The audio recording (if your vendor retains it), the original AI-generated note before any edits, any subsequent edits or addenda, any communication with the patient about the error. Don't delete, don't edit, don't "clean up."
• Don't communicate further with the patient about the matter without your carrier's guidance. Carriers have specific processes for patient communication during a claim; freelancing here can hurt your defense.

Within the first week:

• Retain a copy of your contract with the AI scribe vendor. The carrier may want to review the vendor's contractual obligations and your indemnification rights.
• If your vendor has a process for retrieving the original audio and the AI's original output, initiate that process. The 60-day retention window some carriers recommend exists for exactly this scenario; if you're past 60 days, the original recording may be gone, which makes defense harder.
• Document your standard review-and-sign workflow. The carrier will want to demonstrate that you have a robust review process, not a rubber-stamp process. If you can show "I typically spend 30 to 60 seconds reviewing each note, with extra time on medication-containing sections," that becomes part of your defense.

Long-term:

• The carrier may ask you to commit to a remediation plan. The plan typically involves more explicit verification of high-stakes sections, retention of additional audit trail, or documentation that you've reviewed your AI scribe vendor's accuracy reports. Comply with the remediation plan.
• Don't switch AI scribe vendors mid-claim. The chart is what it is; switching vendors doesn't help and can complicate the timeline if questions arise later.

Vendor contracts: what to look for

Your AI scribe vendor's contract affects your liability exposure. Things to check before signing:

Indemnification. Does the vendor indemnify you against claims arising from technology failures? Most vendor contracts in 2026 limit indemnification to "technology defects in our software" and explicitly exclude clinical use. This is roughly fair (the vendor can't take responsibility for clinical decisions), but you should know what's covered and what isn't.

Data retention. What does the vendor retain, for how long, and is it accessible during a claim? A vendor that deletes audio after 7 days won't have it available 6 months later when a claim surfaces. Many carriers now recommend retention of at least 60 days, and some practices configure vendor retention to match.

Audit trail access. Can you, post-hoc, see when each note was reviewed, what was edited, what the original AI output was? This audit trail is your defense's friend. Vendors that don't provide it leave you with weaker documentation of your review process.

Vendor's own liability insurance. Does the vendor carry errors and omissions insurance, cyber liability insurance, and at what limits? A vendor with no insurance is a bigger risk for you, because they have less capacity to defend joint claims or to contribute to settlement if their tool was involved in patient harm.

Termination and data return. What happens to your data if you cancel? Some vendor contracts have clauses about retention after termination that affect your access to records during a claim filed after you've switched vendors. Check.

A vendor that won't answer these questions clearly, or whose contract has unfavorable terms, is a higher-risk choice regardless of how good the technology is. The best clinical AI in the world doesn't help you when you can't access the audit trail during a claim.

Premium impact in 2026

The current picture, as best as can be determined from carrier reports and broker conversations through early 2026:

• No US carrier known to surcharge for AI scribe use when used per their guidance
• No US carrier known to discount for AI scribe use either, despite arguments that better-documented charts could reduce claims
• Some carriers offer favorable underwriting consideration if the practice can demonstrate good adoption practices (consent, review workflow, audit trail)
• CMPA in Canada does not differentiate fees based on technology use; the fee structure is by specialty and province

The premium-neutral stance reflects the carriers' position that AI scribes are an acceptable but not transformative tool. They neither increase nor meaningfully decrease the practice's risk profile when used responsibly.

This may change as more claim data accumulates. If the data shows AI-scribed practices have lower claim rates (plausible, given the better documentation), discounts may appear. If the data shows higher claim rates from specific error patterns, surcharges might. Neither has happened in the public data through early 2026.

What the carriers want you to do

Synthesized across carriers, the consensus advice for physicians using AI scribes:

1. Review and sign every note. Don't rubber-stamp. The review is your legal protection.
2. Document patient consent. Verbal consent acknowledged in the chart is the minimum.
3. Maintain audit trails. Who reviewed, when, what was edited. The vendor should provide this; if they don't, that's a vendor selection issue.
4. Retain the audio for at least 60 days. This is the carriers' most consistent recommendation that AI scribe vendors don't always default to. Configure it explicitly.
5. Disclose to your carrier. A one-page summary at adoption time, and an annual reminder at renewal.
6. Don't use AI scribes for parts of practice they're not designed for. Procedure notes, complex consultations, and visits with multiple speakers may have different accuracy profiles. Use the tool where it works well; type the rest.

Practices that follow this list have, to the carriers' knowledge, had no measurably different claim experience than non-AI-scribed practices. Practices that don't follow it (sign without reviewing, no consent, no audit trail) have shown up in claims as expected, but the AI-generated note has not typically been the proximate cause of harm so much as an additional layer in the chart.

The honest framing for risk-averse physicians

If you're a physician who chooses AI scribe adoption decisions primarily on the basis of liability risk, the honest answer in 2026 is: the risk profile is comparable to typed documentation, the carriers don't object, and the workflow changes that matter (review, consent, audit trail) are achievable with any major platform.

If you're risk-averse enough that you'd prefer to wait until liability questions are more settled, waiting another 1 to 2 years is a reasonable choice. The technology is advancing, the carrier positions are crystallizing, and the case law is accumulating. By 2027 to 2028, the framework will be even clearer.

If you're risk-averse but want the documentation time savings now, the safe path is: choose a vendor with strong audit trail and contractual clarity, follow the carriers' adoption advice, disclose to your carrier, and document your review workflow. Done that way, AI scribe adoption is well within the boundaries of defensible practice.

---

For the specific case of medication errors in AI-generated notes and how to catch them before signing, see what to do when your AI scribe mishears a medication. For the broader review-and-approve workflow, review and approve AI clinical notes covers the mechanics.