Abridge
Abridge is an enterprise ambient AI platform that turns the clinician patient conversation into a structured, billable note inside the EHR, sold to health systems rather than to individual clinicians. It partners with more than 300 health systems, including Kaiser Permanente, Mayo Clinic, Johns Hopkins, Duke Health, Memorial Sloan Kettering, Northwell, UPMC, Sutter Health and NYU Langone, and won the KLAS Research award for Ambient AI in 2025 and 2026. Its deepest integration is with Epic, where it records in Haiku and places the generated note in the patient chart in Hyperdrive, and it is also integrated with Oracle Health and athenahealth.
It spans outpatient, inpatient and emergency settings, with product lines for clinicians, nursing and revenue cycle. Linked Evidence maps every part of a generated note back to the source audio so a clinician can check any sentence before signing. Through 2026 the platform has added pre visit and pre admission synthesis, Clinical Decision Support that surfaces evidence from UpToDate, the New England Journal of Medicine and JAMA during the encounter, Care Signals, structured order capture, level of service coding recommendations and real time prior authorization with Highmark Health and Allegheny Health Network.
Abridge acquired Altrina in July 2026 to add computer use agents for workflow automation inside the EHR. It was founded in 2018 by Shiv Rao, a practicing cardiologist, and Zachary Lipton of Carnegie Mellon.
Capability Axes
Abridge is the enterprise ambient documentation platform that turns the clinician patient conversation into a structured, billable note inside the electronic health record, sold to health systems rather than to individual clinicians. The AI Health Index grades it on the same fifteen capability axes it applies to every vendor, and the record separates into two halves. On what the product does, the grades are the strongest in its category: A on AI Centrality, A on Clinical and Operational Evidence, A on Autonomy and Oversight Model and A on EHR and Interoperability Depth. On what stands behind the product the grades sit lower, at C on AI Liability and Recourse and C on Commercial Transparency, and the notes on each axis record why. Verified as of Sep 13, 2026.
An AI Health Index grade measures what a buyer can verify from public sources on the date shown. It is not a rating of how good the product is. A vendor can build an excellent system and grade low on an axis because it publishes nothing an outsider can check. How grades read
Abridge's core product listens to the clinical conversation and produces a structured, specialty specific note for the clinician to review inside the EHR. Everything it has added since is model output as well: pre visit and pre admission synthesis, clinical decision support drawn into the encounter, care gap identification, order capture and level of service coding recommendations.
The Altrina computer use agents acquired in July 2026 add model driven workflow execution inside the EHR, and the August 2026 release extended that context aware intelligence to every clinician at partner health systems.
Each note is a draft until the clinician signs it. A structured, specialty specific note is ready for review the moment the visit ends, and the clinician reviews and edits it before finalizing, with editing tools that adapt to individual preferences.
Linked Evidence is the oversight mechanism. Every part of a generated note maps back to the source audio, so a clinician can check any sentence against what was actually said before signing, and Abridge describes it as giving complete auditability for the generated summaries inside Epic. Review becomes something a clinician can audit line by line rather than an act of trust.
The same pattern carries into orders. In the outpatient orders pilot announced in June 2025, medications discussed during the encounter surface inside Epic so the clinician can place orders based on the conversation, with imaging, labs and other order types named as coming next. The system queues the orders, and the clinician places them.
Documentation covers the supply chain and the system architecture rather than the model itself. OpenAI is named on the public subprocessor list alongside Google Cloud, Zendesk, Amplitude and Sentry, so a health system can see which model provider sits in the stack and trace the dependency and the data flow to it.
A security whitepaper and a network diagram are available as documents, and AI security and AI monitoring are cataloged as standing items on the trust center.
The model layer is not described. There is no model card, no architecture description, no statement of what is proprietary and what is a wrapped foundation model, no version or update policy, and no published performance figure. The published clinical evidence describes outcomes at Kaiser Permanente and elsewhere, while the system that produced them carries no published characterization.
The model provider is named outright. The public subprocessor list on the trust center enumerates each party individually, with OpenAI alongside Google Cloud, Zendesk, Amplitude and Sentry, so a compliance team can see that content derived from the encounter reaches a third party foundation model provider.
The surrounding documents fill in the rest of the chain. A third party notices page is published separately, a security whitepaper and a network diagram are available as documents, and the clinician terms incorporate a business associate agreement by reference.
Whether encounter audio or transcripts are retained at the model provider, or used to train or improve models, is not published.
Kaiser Permanente's evaluation of Abridge is published as peer reviewed research in NEJM AI. Over six months and more than 1,000 clinicians, Kaiser Permanente stress tested the platform before deciding to deploy it across the enterprise, in all care settings, specialties, languages and workflows. In the findings Abridge highlights, 1,306 users left 8,986 star ratings, 1,252 self reviews and 3,600 free text comments, and optimized PDQI self reviews averaged 4.35 out of 5.
Abridge supported more than 4 million patient encounters as of November 2024.
KLAS Research gave Abridge its Ambient AI award in 2025 and again in 2026, based on independent interview feedback from customer health systems.
The trust center maps how encounter data moves, cataloging data into the system, data out of the system, breach notification, access monitoring, data asset classification, certificates of destruction, audit logging, multi factor authentication and single sign on.
Every data service vendor in the stack is named: Zendesk, Amplitude, Sentry, OpenAI and Google Cloud. The OpenAI listing indicates that content derived from the encounter reaches a third party model provider. A patient facing explainer is published alongside the clinician guide.
Retention and training use are left out of public material. Nothing published states whether encounter audio or transcripts are retained, for how long, or whether customer content is used to train or improve models, and nothing is published on guardrails, hallucination mitigation or how a safety event is handled.
HIPAA is a named compliance program on the public trust center, next to SOC 2 Type 2, CCPA and TX-RAMP. Supporting documentation is available on request, a master services agreement is among the listed legal documents, and data breach notification is cataloged as its own item. The clinician terms incorporate a business associate agreement by reference.
Abridge states that it applies HIPAA security standards to all data it collects, including data the rule would not otherwise cover, and that a user can delete their account at any time. The enterprise section of the clinicians platform page repeats the HIPAA commitment.
The agreement itself, with its terms and execution path, is not public.
A full public trust center runs on SafeBase. Its reports include SOC 2 Type 2, examined by an independent third party auditor and scoped to the security and confidentiality trust services criteria, alongside SOC 2 Type 1, HIPAA, CCPA, TX-RAMP and WCAG 2.2 AA. TX-RAMP is a Texas state government cloud authorization, and the WCAG 2.2 AA listing is a published accessibility conformance.
The document library covers what a health system security review works from: a security whitepaper, a network diagram, the SOC 2 report, a business continuity and disaster recovery policy and tabletop exercise records. Structured sections describe application and network penetration testing, code analysis, security monitoring, endpoint detection and response, disk encryption, multi factor authentication, audit logging and cyber insurance.
The subprocessor list is public and names Zendesk, Amplitude, Sentry, OpenAI and Google Cloud, so the third parties in the stack are visible before any review begins. Substantive documents sit behind an access request, while the existence of the SOC 2 report, its scope and the independence of its auditor are all clear without one.
The core product drafts clinical documentation for a clinician to review and sign, and Abridge claims no FDA clearance for it.
The company's statement on the subject sits in Section 11(d) of the clinician terms of use, under a section headed Representations, Warranties and Disclaimer. It says Abridge does not review, monitor, evaluate or analyze patient information for any medical purpose, does not engage in the practice of medicine and does not provide medical advice or diagnoses, and that the clinician is responsible for diagnosis and for the standard of care.
The product has grown since that language was written. Clinical Decision Support is now a named first party capability that surfaces UpToDate, NEJM and JAMA content into the encounter, and the August 2026 release added level of service coding recommendations. The intended purpose language in the terms has not changed since November 25, 2025, so it predates both layers and does not speak to them.
Abridge also lists two conformances outside the FDA framework, TX-RAMP and WCAG 2.2 AA.
A dedicated AI section on the trust center catalogs three items: AI training data and bias, AI security and AI monitoring. Bias in training data is a named category a customer can request documentation on, and model monitoring is presented as a standing control rather than a launch activity. The public subprocessor list, which names OpenAI, lets a health system see the model provider underneath the product.
The contents of the AI section are available on request, and no result is public. There is no bias evaluation output, no demographic or subgroup performance analysis, no error taxonomy and no measured accuracy figure, including for the languages beyond English that Abridge supports.
The clinician terms of use set out how liability is shared, and most of the weight sits with the clinician and the health system.
The platform is offered as is and as available, and warranties of accuracy, reliability, completeness and freedom from error are expressly disclaimed. Aggregate liability is capped at the greater of $1,000 or the fees paid in the previous twelve months. Because the platform is often supplied to an individual clinician at no cost through a health system agreement, the effective cap for that user can be the $1,000 figure.
Indemnification runs one way. The clinician defends and indemnifies Abridge, and no reciprocal obligation is stated. Disputes go to binding arbitration with a class action waiver and a one year limitation period, with a thirty day opt out available by written notice to Abridge's legal team.
A limited performance warranty sits beside the disclaimers. Abridge commits that the platform will be provided in a professional manner consistent with industry standards and will substantially conform in all material respects to the user documentation, though no remedy is attached to a breach of it.
Linked Evidence maps every part of a generated note back to the source audio, so a clinician can find and fix an error before signing. For an error that gets through, the terms offer no error rate, published limitation set, warranty remedy or patient facing recourse route.
The recorded patient is not a party. The terms state that no person other than the parties is entitled to any benefit under them, and correcting a signed note runs through the health system's own amendment process rather than through Abridge.
Two more clauses shift responsibility toward the customer. Obtaining patient consent to recording sits with the clinician, and publishing benchmark results or comparisons of the platform requires Abridge's prior written consent, which limits a customer's ability to publish its own evaluation.
Abridge works inside Epic rather than beside it. The integration is embedded in Epic Haiku and Hyperdrive, so recording, review and the finished note stay in the Epic workflow with no copy and paste step and no parallel interface. The February 2024 Epic announcement described the path as running from recording conversations in Haiku to embedding generated notes in the patient chart in Hyperdrive.
The inpatient release added depth. Epic maps the generated note into whichever note type the clinician selects, SmartPhrases and templates support review and rapid note closure, and the inpatient workflow handles pre charting, post charting and switching between patients. Abridge says it is used across outpatient, emergency department and inpatient settings.
Integration reaches past documentation. Order capture now runs in the background of the conversation, a capability that began in June 2025 as an early pilot in Epic's Workshop program surfacing the medications discussed in an encounter inside Epic. Real time prior authorization workflows run with a payer and delivery system partner. Outside Epic, Abridge is also integrated with Oracle Health and athenahealth.
Abridge runs as a hosted service, the only deployment it publishes, across more than 300 health systems, from academic centers such as Mayo Clinic, Johns Hopkins and Memorial Sloan Kettering to pediatric hospitals such as Seattle Children's and community and safety net providers such as AltaMed and Cambridge Health Alliance.
Residency is committed for the United States. The enterprise section of the clinicians platform page states that all data is stored in secure United States based centers, with 256 bit encryption in transit and at rest, single sign on and governance controls the customer can configure. The trust center covers operational resilience, with a business continuity and disaster recovery policy, an alternate processing and storage site, contingency training, tabletop exercise records, status monitoring and anti DDoS protection.
Whether processing at OpenAI or Google Cloud stays in the United States is not stated; the storage commitment does not by itself cover it.
Distribution is enterprise only. Abridge reaches its 300+ health systems through institutional contracts rather than individual clinician signup, and there is no self serve path.
The trust center lists a master services agreement among its legal documents, held behind an access request alongside the security documentation.
Value is argued through customer case studies, among them a WVU Medicine study of the return on investment from scaling Abridge across 50+ specialties in rural communities. No rate card, per clinician figure, tier structure or contract structure is published.
Coverage spans outpatient, inpatient and emergency care, with dedicated product lines for clinicians, nursing and revenue cycle. The clinicians platform page says Abridge is used at over 40 specialties, with a structured, specialty specific note produced for each visit.
Specialty breadth is backed at named customer sites. Inova runs it across 40+ specialties and 9,400+ clinicians, WVU Medicine across 50+ specialties in rural communities, and Akron Children's across 22 pediatric specialties with 95 percent user retention. Seattle Children's selected it for 18 pediatric specialties, Memorial Sloan Kettering uses it in oncology and Hospital for Special Surgery in orthopedics. In August 2026 Abridge extended its context aware clinical intelligence to every clinician at partner health systems, which widens the user population rather than the specialty list.
Language support is stated less clearly than specialty coverage. The 28 languages figure on the customers page describes AltaMed's multilingual patient population, not the number of languages the platform supports, and the clinicians platform page gives no language count. Abridge's February 2024 Epic announcement named English and over 15 additional languages, including Spanish, French, German, Italian, Chinese, Japanese and Haitian Creole, and put specialty support at over 50.
Citable summary
Self contained paragraphs, free to quote with attribution. Grades shown resolve from this record and change when it is regraded.
Epic integration and HIPAA posture, answered from the record
Abridge is embedded natively across the Epic surfaces Haiku, Canto and Hyperdrive with bidirectional note sync rather than copy and paste, alongside a partnership with athenahealth, and the AI Health Index grades it A on EHR and Interoperability Depth, recording the integration as the deepest in the index. On HIPAA the honest framing is that no body issues a HIPAA certificate, so compliance is established through artifacts rather than a badge. The artifacts on record: HIPAA is a named program on the public trust center, Abridge states that it applies HIPAA security standards to all data it collects, breach notification is separately cataloged, and the business associate agreement is incorporated by reference into the published clinician terms. The AI Health Index grades it B on HIPAA and BAA Posture, held below the top grade because the agreement itself, its tier and its execution path are not public, so the shape of the arrangement has to be established in the sales conversation. Verified as of Sep 13, 2026.
Source: AI Health Index, Sep 13, 2026
The model provenance disclosure the rest of the category avoids
Abridge names its model provider. The public subprocessor list enumerates OpenAI alongside Google Cloud, Zendesk, Amplitude and Sentry, so a buyer can see that encounter derived content reaches a third party foundation model provider rather than inferring it from a proprietary artificial intelligence claim, and the AI Health Index grades that A on Model Supply Chain Disclosure, the specific artifact it recommends asking every scribe vendor for. The naming makes two further questions possible, and the index records both as open: whether each named party is individually covered by a business associate agreement, and whether encounter audio or transcripts are retained at the model provider or used to train or improve models. The counterpart finding sits on liability. Linked Evidence maps every part of a generated note back to the source audio, the most rigorous oversight mechanism the AI Health Index has graded in this category, while the published terms disclaim accuracy and attach no remedy to the performance warranty, which is why oversight grades A and AI Liability and Recourse grades C on the same record.
Source: AI Health Index, Sep 13, 2026
Common questions
Does Abridge integrate with Epic and is it HIPAA compliant?
On Epic the answer is unambiguous. Abridge is embedded natively across Haiku, Canto and Hyperdrive with bidirectional note sync rather than copy and paste, extends beyond documentation into structured orders and real time prior authorization workflows, and also carries an athenahealth partnership. The AI Health Index grades it A on EHR and Interoperability Depth and records the integration as the deepest in the index. On HIPAA the precise answer matters: no body issues a HIPAA certificate, so HIPAA compliant is a self description for any vendor, and what a buyer can verify are the artifacts. Abridge lists HIPAA as a named program on its public trust center, states that it applies HIPAA security standards to all data it collects, catalogs breach notification separately, and incorporates the business associate agreement by reference into its published clinician terms. Its installed base of major academic and integrated health systems also demonstrates that executed agreements exist at scale, since institutions of that kind do not deploy an ambient recording product without one. The AI Health Index grades it B on HIPAA and BAA Posture as of Sep 13, 2026, held below the top grade because the agreement terms, tier and execution path are not published.
Does Abridge send patient data to OpenAI?
Abridge is one of the few ambient scribe vendors whose public disclosure lets a buyer answer this question at all. Its subprocessor list names OpenAI alongside Google Cloud, Zendesk, Amplitude and Sentry, so encounter derived content demonstrably reaches a third party foundation model provider, and the dependency can be diligenced rather than inferred. That enumeration is what earns A on Model Supply Chain Disclosure from the AI Health Index, which treats it as the artifact to request from every vendor in the category. What the disclosure does not state, and what the AI Health Index records as the two questions to put to the vendor directly: whether each named subprocessor is individually covered by a business associate agreement, and whether encounter audio or transcripts are retained at the model provider or used to train or improve models.
What happens when an Abridge note is wrong?
Before signing, the correction machinery is the strongest the AI Health Index has graded in this category. Linked Evidence maps every part of a generated note back to the source audio, so a clinician can verify any sentence against what was actually said, which converts review from a trust exercise into an auditable one and earns A on Autonomy and Oversight Model. After signing, the published terms govern, and they read differently: the platform is offered as is with accuracy expressly disclaimed, aggregate liability is capped at a level that can fall to a small fixed floor where the product reaches a clinician at no cost through a health system agreement, indemnification runs from the clinician toward the vendor rather than the reverse, and the terms state that no person other than the parties benefits under them, so the recorded patient is expressly not a beneficiary of the contract governing the system that captured the encounter. Correcting a signed note runs through the health system amendment process rather than the vendor. The AI Health Index grades this C on AI Liability and Recourse, and the grade records the distance between excellent means of catching an error before harm and no commitment standing behind one that gets through.
Does Abridge publish pricing?
No. The AI Health Index grades Abridge C on Commercial Transparency as of Sep 13, 2026. Distribution is enterprise only, through institutional contracts rather than individual signup, and while a master services agreement is cataloged among the legal documents on the public trust center, it sits behind an access request alongside the security documentation. The contrast inside the category is the useful fact for a buyer: several smaller ambient scribe vendors publish per clinician rates openly, so gated pricing at the category leader is a choice rather than a necessity of scale, and neither the price nor a published return figure can be established from public material before a sales conversation.
Does Abridge pay to be listed on the AI Health Index?
No. The AI Health Index is researched from public sources, no vendor pays for inclusion, for a grade or for placement, and every record carries the date it was last verified. A vendor that publishes more is regraded and the change is logged.
Product updates
Material product, regulatory, evidence and commercial changes at Abridge, each verified against a live source and tagged to the capability axis it bears on. Funding rounds and awards are not product changes and are not logged.
Abridge launched pre bill review for clinical documentation integrity, coding and revenue cycle teams. It checks inpatient claims before submission, comparing the coded diagnoses and DRG against what the clinical record supports, including what was captured in the encounter, and surfaces discrepancies with the evidence. Abridge does not change documentation, codes or claim status; the CDI team decides what to hold, correct or release.
Abridge shipped its August 2026 release with four additions that push the product upstream of the encounter it was built to document. Pre Visit Summaries synthesize a patient's history, active problems and recent changes before the visit, with every detail linked back to its source in the chart. Pre Admission Summaries do the equivalent for inpatient handoff, aggregating the emergency department course, labs, vitals, imaging and prior documentation into one view at the start of an admission. The release also adds transparent level of service recommendations for coding, and lets clinicians claim AMA PRA Category 1 CME credit for reviewing eligible topics at the point of care.
Abridge announced the enterprise-wide expansion of its Context-Aware Clinical Intelligence platform, allowing partner health systems to deploy the AI agent to every clinician across their organizations.
Alternatives to Abridge
Vendors in the same categories as Abridge whose graded profiles look most alike, with a line on what each product does.
Similar is not the same, so check that each one does the job you need. No vendor pays for placement.
CarePilot sells AI documentation, coding and order drafting to independent and community practices that run athenahealth.
Microsoft Dragon Copilot is the ambient clinical documentation assistant Microsoft created in March 2025 by merging DAX Copilot with Dragon Medical One, both from its 2022 acquisition of Nuance.
Nabla sells an ambient AI assistant that listens to the visit and drafts a structured clinical note, together with dictation and E/M coding suggestions, to health systems, medical groups…
voize is a German voice documentation company built for nurses and caregivers rather than physicians, with its speech model running on the caregiver's device.
Ambience Healthcare sells an AI platform for documentation, coding and clinical documentation integrity to large health systems and academic medical centers.
Empathia AI is a general purpose ambient scribe built deliberately for two markets at once, claiming compliance with HIPAA in the United States and PHIPA and PIPEDA in Canada with tailored…
Ordered by how closely the graded profiles match across the fifteen axes, inside the same categories. See how each axis is graded in the methodology.
Compared With
Each comparison carries a written verdict, the buyer conditions that favor each vendor, and a graded side by side. Pairs that cross a category boundary are grouped separately, and their verdicts state where the boundary sits rather than manufacturing a head to head.
Head to head
Vendors the index assesses as direct competitors to Abridge for the same buyer.
Adjacent comparisons
Products a buyer researches alongside Abridge that do a different job: a different category, a different layer of the stack, or a specialist scope. These pages exist to settle whether the comparison is real before it settles which one to pick.
Announced Deployments
Publicly announced health system deployments and partnerships. This is a record of announcements, not an assessment of deployment success or scale.
Pricing
Vendor-published figures are labeled as such. Figures labeled “Estimated” are derived from third-party sources and have not been confirmed by the vendor.
| Entry Price | Pricing Basis | BAA Tier | Implementation | Source |
|---|---|---|---|---|
|
Not published. Enterprise contracts negotiated per health system
|
Enterprise contracts negotiated per health system, with no self serve path | — | — | Vendor Published |
Abridge sells to health systems on negotiated contracts, with a master services agreement among the legal documents in its trust center, behind an access request, and argues value through customer case studies, including a WVU Medicine return on investment study across more than 50 specialties. It publishes no pricing.