Scarlet Series B Puts Faster AI Medical Device Certification Against the Limits of Regulatory Scale

Scarlet raised a $50 million Series B to expand AI medical device certification, confronting a bottleneck that conventional review systems struggle to handle. The Scarlet Series B was led by Headline and announced on October 12, 2026. Notion Capital joined as a new investor, while Atomico, Creandum, Kindred Capital, and Entrepreneurs First also participated.

The financing is significant because Scarlet is not selling another compliance dashboard. It operates as a government-designated certification body whose reviewers can make consequential decisions about regulated medical devices. The company plans to expand in the United States and Asia-Pacific while hiring more expert reviewers and engineers.

That makes this a test of a difficult proposition. Scarlet wants to increase certification speed without turning regulatory judgment into automated box-checking. Established certification bodies have scale, history, and broad coverage. Scarlet argues that specialization, software-assisted evidence processing, and continuous oversight fit AI products better.

The money can fund expansion, but it cannot eliminate the tension between speed and scrutiny. Scarlet must show that its operating model scales across jurisdictions while preserving independent, traceable, and technically informed decisions.

What the Scarlet Series B Actually Funds

The Scarlet Series B finances regulatory capacity, not a broader grant of regulatory authority.

Scarlet’s official funding announcement says the company will use the capital to grow its expert team. It also plans to increase its footprint in the US and Asia-Pacific and widen the range of AI it can assess.

Those goals require more than opening regional offices. Medical device rules differ across markets, and each designation carries a defined legal scope. Expansion therefore depends on regulatory recognition, qualified personnel, operating controls, and the ability to support local manufacturers.

The company was founded in London by James Dewar and Jamie Cox after both worked on medical AI at Babylon Health. Scarlet spent its early years securing regulatory designations before issuing certificates. That history matters because certification authority cannot be created merely by releasing software or hiring a sales team.

The round’s investor group reflects that long timeline. Headline led the financing, while Notion Capital became a new investor. Existing backers Atomico, Creandum, Kindred Capital, and Entrepreneurs First returned. Headline partner Nancy Xiao Mclaughlin also joined Scarlet’s board.

According to the reported funding details, Scarlet’s customers upload technical documentation through its platform. Qualified reviewers then assess the evidence against the rules governing the relevant device and market.

The software handles document intake, organization, and traceability. Human reviewers retain responsibility for regulatory judgments. This division is central to Scarlet’s pitch because certification involves interpretation, risk assessment, and evidence evaluation.

Scarlet says customers subscribe for ongoing surveillance, annual audits, and product-change assessments. That model treats certification as a continuing relationship rather than a one-time gate before market entry.

This approach is particularly relevant to software as a medical device, or SaMD. SaMD is software that performs a medical function without being part of a physical medical device. Its behavior can change through model updates, data pipelines, interface revisions, or deployment changes.

A conventional certification project often focuses on a defined product version. AI developers, however, want to revise models and workflows more frequently. Every material change can raise questions about safety, effectiveness, documentation, and the need for another regulatory submission.

Scarlet says its customers have shortened some product-update cycles from months to weeks. It also claims customers have removed a year or more from certain certification timelines. These figures are company-reported outcomes, not independent benchmarks across all device classes.

The distinction is important. A financing announcement does not establish that every medical device can follow the same accelerated path. Device risk, clinical evidence, software architecture, and submission quality all affect review time.

The funding gives Scarlet resources to test its model at greater scale. It does not settle whether the model will produce consistent results across more products, reviewers, and regulatory systems.

Why AI Medical Device Certification Has Become the Bottleneck

Medical AI can change faster than the institutions responsible for evaluating it, but faster review cannot substitute for sufficient evidence.

Scarlet and its investors describe a widening mismatch between software development and medical device certification. A team can train or revise a model within weeks. The resulting device can then spend many months in assessment, especially when documentation is incomplete or reviewers lack specialized capacity.

The funding report says an expert assessment can take between 12 and 18 months. It adds that approval of a minor update can face another six to eight months of delay. Those estimates describe the problem Scarlet wants to address, rather than a universal timetable.

Demand is also increasing. The FDA says it had authorized more than 1,600 AI-enabled medical devices by September 2026. Its AI device landscape includes diagnostic tools, image-processing systems, monitoring products, and treatment-support software.

That count creates pressure at several points. Manufacturers must prepare evidence, regulators must maintain policy, and reviewers must understand increasingly complex software. Hospitals and clinicians also need assurance that authorized systems remain safe after deployment.

AI products make this work harder because their risk is not confined to source code. Performance can depend on training data, input quality, clinical context, user behavior, infrastructure, and the population receiving care.

Some systems also change after launch. The FDA permits certain AI device submissions to include a predetermined change control plan. A PCCP describes planned modifications and the methods used to develop, validate, and implement them safely.

That framework supports controlled iteration, but it does not create unlimited permission to update a product. Changes outside an authorized plan can still require review. Manufacturers must also monitor whether real-world performance remains consistent with the evidence supplied before launch.

Scarlet’s opportunity sits inside this lifecycle problem. If documentation remains structured, findings remain traceable, and reviewers can assess changes efficiently, manufacturers can avoid rebuilding every submission from scattered files.

The harder issue is whether a certification body can preserve context across repeated changes. A model update can affect clinical performance, cybersecurity, human factors, and risk controls. The review must identify which assumptions remain valid and which evidence needs to change.

Scarlet says its platform ties findings to supporting sources and regulatory clauses. That design can reduce time spent searching through submissions. It can also make review decisions easier for regulators and manufacturers to audit.

Yet administrative efficiency and evidentiary sufficiency are different outcomes. Organizing a weak clinical study does not make the study stronger. Detecting a missing document quickly does not resolve the missing evidence.

The strongest version of Scarlet’s case is therefore narrower than “AI makes certification fast.” Its model can remove avoidable friction while leaving safety judgments with qualified experts.

This distinction should matter to medical AI developers. A faster reviewer cannot compensate for an unclear intended use, unsuitable validation data, or weak monitoring controls. Better preparation and faster feedback must operate together.

The market pressure is likely to increase as generative models enter medical workflows. Generative systems can produce variable outputs, making their failure modes different from those of deterministic software. Reviewers need methods that address output variability, clinical consequences, and changing model behavior.

For Scarlet, that demand represents a larger addressable market. It also raises the technical and institutional burden behind each certification decision.

Scarlet Versus the Generalist Certification Model

Scarlet is betting that specialization can outperform generalist certification without weakening the independence that gives certification its value.

Traditional notified bodies assess many types of medical devices. Their portfolios can include implants, diagnostic equipment, hospital hardware, laboratory products, and software. That breadth supports scale, but it can also make specialist software expertise difficult to allocate.

Scarlet positions itself around AI and software medical devices. The company says it is the only European notified body dedicated to those categories. Its argument is that focused reviewers and software-aware processes can evaluate iterative products more consistently.

The primary opponent is not a single competitor. It is the project-based, document-heavy certification model built around products that change less frequently than software.

Headline’s investor thesis describes legacy reviews as inconsistent, opaque, and poorly suited to continuous software development. Those claims reflect the lead investor’s perspective and should not be treated as an independent comparison.

The criticism still identifies a real design question. Medical software teams often organize evidence across issue trackers, quality systems, clinical studies, model records, and risk documents. Reviewers need to connect those materials to specific regulatory requirements.

Scarlet’s platform reportedly accepts documentation in multiple formats and checks whether submissions are complete. It then organizes evidence so qualified reviewers can evaluate the relevant claims and controls.

This process can improve consistency when it standardizes how findings are recorded. It can also provide manufacturers with clearer explanations about what must change. Both benefits depend on the quality of the underlying review rules and reviewer judgment.

The subscription model creates another difference. Instead of treating every product modification as a separate engagement, Scarlet can maintain an ongoing view of the manufacturer and device. Continuous surveillance and annual audits remain part of that relationship.

Flok Health illustrates the intended use case. The company operates an AI-supported physiotherapy service and wants to release medical device upgrades twice each month. Its co-founder, Finn Stevenson, said Scarlet’s approach allows the company to pursue that cadence.

That statement shows the commercial appeal of faster certification. It does not mean every twice-monthly release requires or receives a new certificate. The regulatory treatment depends on the nature and significance of each change.

A continuing relationship can help both sides classify those changes. Reviewers can understand the existing device, while the manufacturer can raise proposed modifications earlier. That should reduce repeated orientation work.

However, ongoing commercial relationships also require strong safeguards. A certification body receives revenue from manufacturers while remaining responsible for independent assessment. That structural tension exists across conformity assessment, not only at Scarlet.

Specialization can strengthen independence when reviewers understand the technology well enough to challenge manufacturers. It can weaken confidence if commercial speed becomes the dominant performance measure.

Scarlet therefore needs evidence that its model is both faster and appropriately demanding. Useful measures would include review-cycle duration, the number of evidence rounds, audit findings, post-market issues, and regulator feedback.

Scale will complicate that proof. A small specialist team can maintain shared standards through close collaboration. A larger international workforce needs formal training, calibration, quality assurance, and escalation paths.

The company also faces established organizations with broader resources and existing relationships. Generalist bodies can hire software specialists, build digital review systems, and introduce recurring services of their own.

Scarlet’s defensibility rests partly on regulatory designations that take years to secure. It also depends on operational knowledge gained from repeated AI and software assessments.

Neither advantage is permanent. Designations establish permission to operate within a defined scope. They do not guarantee superior review quality, customer outcomes, or regulatory acceptance in every market.

US Recognition Does Not Mean Universal Approval

Scarlet’s international expansion will succeed only if customers understand the limits of each authorization and review pathway.

Scarlet is recognized by the FDA as a third-party review organization under the 510(k) program. The agency’s recognized reviewer list names Scarlet NB B.V. and identifies eligible device categories.

The scope includes selected devices across areas such as anesthesiology, cardiovascular medicine, neurology, pathology, and radiology. However, the recognition does not let Scarlet clear every medical device or act as the FDA itself.

Under the third-party program, recognized organizations can review eligible 510(k) submissions and make recommendations. The FDA retains its regulatory role, and eligibility depends on the applicable device product code.

That distinction matters because “FDA recognized” can sound broader than it is. Manufacturers must confirm whether their device qualifies for third-party review and whether Scarlet covers the relevant category.

European certification follows another legal structure. A notified body performs conformity assessment for devices requiring external review under the European Union’s Medical Device Regulation. Its designation defines which device types and conformity activities it can assess.

The United Kingdom operates its own framework, with approved bodies handling defined conformity assessment functions. Asia-Pacific is not one regulatory market at all. It includes national systems with different evidence, registration, and local representation requirements.

Scarlet cannot copy one European process into every country. It needs jurisdiction-specific expertise, regulatory relationships, and operating entities that meet local rules.

The company’s expansion plan therefore contains two connected challenges. It must build review infrastructure, and it must obtain or maintain the authority required for each service.

Even within an existing designation, regulators can monitor a certification body’s work. A Dutch oversight record documented surveillance activity involving Scarlet in 2024. It also reflected the organization’s early stage and limited assessment volume at that time.

This kind of supervision is not a warning sign by itself. Oversight is part of the system that gives notified-body decisions legal and clinical credibility. It also shows why expansion cannot rely only on software efficiency.

Regulators need confidence in reviewer competence, impartiality, documentation, audit controls, and management systems. Hiring quickly can create pressure on each of those elements.

Scarlet must also manage different definitions of acceptable change. An update considered minor in one framework might trigger additional review in another. A model used for decision support can face different requirements when its intended use changes.

Generative AI adds further uncertainty. A system that summarizes records presents different hazards from software that recommends treatment. The consequences of an incorrect output also influence the required evidence and controls.

Scarlet says it wants to certify AI across healthcare in all its forms. That ambition should be read as a business direction, not proof that every form already sits within its present regulatory scope.

Customers should ask precise questions before relying on the global message. Which Scarlet entity handles the review? Which authority designated it? Which device codes and risk classes fall within scope? What action does the relevant regulator still need to take?

Those questions do not diminish the company’s role. They define it accurately. Regulatory specificity will become more important as Scarlet enters markets where the brand is better known than its formal authorizations.

The Real Tradeoff Is Speed Versus Verifiable Judgment

Scarlet must prove that faster certification comes from removing wasted work, not compressing the judgment that protects patients.

The company’s strongest claim is that delay does not automatically produce rigor. Few manufacturers would disagree. Waiting for reviewer availability, searching fragmented documents, or receiving unclear feedback does not make a medical device safer.

Software can reduce those delays. It can detect missing materials, map evidence to requirements, maintain version histories, and show how a finding was resolved. These are practical improvements to regulatory operations.

They also create measurable outputs. Scarlet can track how long submissions remain incomplete, where reviews pause, and which requirements generate repeated questions. That information can help manufacturers prepare stronger evidence earlier.

The risk begins when speed becomes a substitute measure for quality. Certification quality is harder to summarize than turnaround time. A quick decision can be correct, but speed alone cannot demonstrate that result.

Scarlet’s customers and investors have incentives to highlight shorter timelines. Medical device companies want market access and more frequent releases. Investors want a scalable business with expanding demand.

Regulators and patients have a different priority. They need confidence that evidence supports the device’s intended use and that risks remain controlled after deployment.

Qualified reviewers sit between those incentives. Scarlet says humans make the final regulatory judgments while technology processes documentation. Maintaining that boundary is essential.

Automation can rank, extract, and connect evidence, but it can also propagate errors. A system may associate a study with the wrong product version or overlook a limitation hidden in an appendix.

Human reviewers must understand how automated tools produced their outputs. They also need time to investigate contradictions rather than accepting a neatly organized record.

AI-assisted certification introduces another layer of governance. Scarlet must validate its internal software, control changes to that software, and document where automation influences reviewer work.

Review consistency is valuable, but uniformity can become a weakness when unusual devices require deeper analysis. A standardized workflow should surface exceptions rather than force them into familiar categories.

International growth increases this pressure. New reviewers bring different experience, and new markets introduce different regulatory interpretations. Scarlet needs calibration processes that produce consistent reasoning without suppressing appropriate professional disagreement.

The public evidence remains limited. Scarlet has disclosed customer examples and claimed substantial timeline reductions. It has not published a broad, independently audited comparison against other certification bodies.

That absence does not invalidate the model. It means the financing thesis is ahead of the available performance record.

A credible proof package would separate administrative time from substantive review time. It would show whether faster intake also reduces total certification duration. It would account for device risk, submission readiness, and change complexity.

Post-market outcomes matter too. Faster certification looks less valuable if manufacturers later face corrective actions, safety concerns, or unexpected review work. Conversely, stable real-world performance would strengthen Scarlet’s case.

Regulatory trust compounds slowly. A serious failure can damage it quickly. Scarlet’s expansion strategy must therefore treat review quality as the constraint around growth, not as a metric to optimize later.

The company also needs to avoid overpromising what certification provides. Authorization supports market access after a defined assessment. It does not guarantee clinical adoption, reimbursement, integration, or strong performance in every hospital.

AI systems can encounter population shifts, workflow differences, and new data conditions after launch. Certification is one part of responsible deployment, alongside monitoring, clinical governance, cybersecurity, and incident response.

Scarlet’s subscription structure might support that lifecycle perspective. Continuous engagement can keep the certification body informed as the device changes. Its value depends on whether surveillance identifies meaningful risks before they become patient harm.

The Scarlet Series B is therefore a bet on institutional design. The software matters, but the defensible product is trusted judgment delivered with less avoidable friction.

Three Signals Will Show Whether Scarlet Can Scale

The next stage should be judged through regulatory scope, measurable review performance, and evidence that quality survives international growth.

The first signal is concrete regulatory progress in the US and Asia-Pacific. Scarlet should identify the jurisdictions, device categories, and review pathways it can support. New offices or hires matter less than verified authority and completed assessments.

Broader FDA eligibility would strengthen the expansion thesis if it comes with satisfactory performance. Additional recognition in Asia-Pacific markets would show that Scarlet can translate its operating model into distinct legal systems.

A lack of new authorizations would weaken the global story. It would suggest the company remains primarily a European specialist with selected US capabilities, despite wider commercial ambitions.

The second signal is transparent operational evidence. Scarlet should disclose review times with enough context to make comparisons meaningful. Metrics should distinguish initial completeness checks, substantive assessment, manufacturer response periods, and final decisions.

Customer claims about saving a year are attention-grabbing, but they do not establish a typical result. Median timelines, device categories, and evidence-round counts would offer a clearer picture.

The company should also report how often product changes proceed under existing controls. That information would test whether continuous certification supports real software iteration or mainly improves documentation.

Independent customer case studies would strengthen those numbers. The best examples would explain the device, risk class, original process, Scarlet workflow, and final regulatory outcome.

The third signal is quality under scale. Scarlet plans to hire more reviewers and engineers, but headcount alone cannot prove capacity. Training, calibration, audit findings, and post-market performance will provide better evidence.

Regulatory monitoring will be especially important. Favorable surveillance results would support the claim that Scarlet can grow without lowering assessment standards. Corrective findings would reveal where the operating model needs more control.

Established certification bodies will also respond. They can develop software-focused teams, automate document workflows, and offer more continuous engagement. Their response will test whether Scarlet owns a distinct category or simply moved first.

Medical AI developers should watch those signals closely. Certification affects release planning, investment timing, clinical partnerships, and the cost of maintaining products across markets.

Healthcare buyers should care for a different reason. Faster market entry increases the number and frequency of AI products requiring local evaluation. Buyers still need evidence about clinical fit, monitoring, and accountability after certification.

The $50 million round gives Scarlet time and resources to build the institution it describes. It also raises the standard of proof. Investors have funded an international expansion, not merely a promising workflow.

The Scarlet Series B will look justified if the company earns broader regulatory scope, publishes credible performance data, and maintains review quality as volume rises. Without those signals, faster certification will remain an attractive claim rather than a demonstrated institutional advantage.

For teams building regulated medical AI, the immediate action is straightforward. Map every planned model change to its evidence, monitoring, and regulatory consequences before release. Then ask whether a continuing certification relationship reduces genuine review friction while preserving independent scrutiny.