Why TORTUS pursued Class IIa device certification 

AVT is a powerful technology, which means both tremendous benefit and potential harm to patients. By our assessment, that makes it a medical device. Only Class IIa carries a level of rigour proportionate to that risk, because it requires external audit, while Class I is self-declared and carries little real assurance. And the technology's trajectory makes certification unavoidable: everything further up or downstream in the clinical workflow carries even higher risk, and has to be built on a device foundation from the start. It cannot be bolted on later.

AVT becomes part of how a clinician thinks

AVT does not transcribe, it composes. It combines an LLM with speech-to-text to take over part of the clinician's cognitive workload, constructing a note from an overheard conversation rather than capturing dictated words. The scale is the risk: an average 11-minute consultation contains over 150 clinical entities, and every one is exposed to word-error rates, hallucinations that inject harmful data, and omissions that remove critical data. Reliance bias compounds this over time rather than resolving it.

Once a note is committed, it is permanent and carries no AI provenance, because EHR data ontologies have no field for it. The error becomes indistinguishable from clinician judgement, forever. The harms are not hypothetical: lawsuits and high-profile hallucinations affecting care are already documented. Neither are the benefits: clinicians report daily that TORTUS surfaced something they had missed. Anything that can directly create or prevent patient harm should be regulated like the medical device it functionally is.

Class IIa is the only certification that proves any of this has been externally validated

Most AVT is self-declared Class I, which in the UK means zero external oversight. Of 19 vendors on the national AVT registry, almost none carry the external audit that advanced certification requires. The “medical device” label is unearned: you can declare a vacuum cleaner a medical device, and absent a challenge or audit, it remains one. The MHRA lacks the resources to police this, so the badge alone gives a buyer no confidence the claims are validated.

Class IIa is categorically different. It requires an externally audited quality management system, hundreds of documents evidencing how every component works and how it is tested and monitored, and mandatory real-world post-market surveillance. To meet that last requirement, we built a monitoring system from scratch that tracks live hallucination rates and other metrics across deployments in real time.

This is not theoretical. There are active US class actions against ambient scribe deployments (Sharp HealthCare and Abridge, November 2025), and a Canadian provincial auditor found serious transcription errors across 20 systems in procurement testing, warning they could produce harmful treatment plans. Malpractice carriers now flag hallucination as a live liability category.

The AVT roadmap moves into higher-risk workflow, and that step change cannot be bolted on later

Documentation is only the first utility of having an AI in the room. The workflow before and after it, summarising the record, generating downstream orders and tasks, demands higher safety thresholds and semi-autonomous operation. Suggesting or placing a prescription carries an order of magnitude more clinical risk than scribing. As these systems become more like colleagues, a vendor without Class IIa hits a ceiling: crossing into regulated territory means 18-plus months and a full rebuild of the entire device under that regime.

You can mix 0.9% saline in your kitchen with a jug and table salt - chemically identical on paper to an IV bag, but night-and-day different in safety assurance, because the assurance is built in from the start and cannot be retrofitted.

AVT is powerful enough to warrant regulation. Regulation means external audit. The roadmap requires that audit regardless. Certifying AVT as Class IIa is not a nice-to-have, it is the only assurance that matches the risk an organisation is actually taking on. As of June 2026, TORTUS is at 2.5m NHS consultations and counting, proof that building it right is no barrier to shipping. In fact, it is quite the opposite.

Read more on how we went on this journey and what it means for organisations thinking of procuring AVT in the blogs below.