What if your team could spot ambiguous requirements the moment they’re written, keep trace links current without manual cross-referencing, and cut review cycles from weeks to days? That’s what AI brings to requirements management in 2026. Tools built on natural language processing (NLP), machine learning (ML), and large language models (LLMs) now give engineers immediate feedback on quality, traceability, and risk, right inside their authoring workflow. The payoff is biggest in regulated industries where a single vague requirement can ripple into months of rework.

This guide covers where AI delivers value today, what the risks and limitations are, how to evaluate tools, and what a real AI-powered requirements workflow looks like.

What Is AI in Requirements Management?

AI in requirements management means applying pattern detection, quality checks, and relationship mapping to the work of writing, tracing, and validating large requirement sets. Engineers derive, decompose, trace, rewrite, and evolve large numbers of engineering artifacts, and that work is time-consuming and prone to human error.

AI changes that by giving engineers immediate feedback. When someone writes “the system shall respond quickly to overcurrent conditions,” AI flags the requirement as unverifiable because there’s no measurable threshold, instead of waiting three months for a test engineer to discover the ambiguity.

Key Technologies Driving AI Requirements Management

Three technologies power most of what you’ll see in AI requirements tools today:

• Natural language processing (NLP): The most mature. Tools already use NLP to check requirements quality against INCOSE and EARS criteria for clarity, completeness, and verifiability.
• Machine learning (ML): Goes beyond rule-based checking to learn from historical data. Traceability is the standout ML application in requirements engineering so far.
• Large language models (LLMs) and predictive analytics: The research frontier. LLMs generate, restructure, and reason over requirements content, while predictive models forecast which requirements carry the highest risk of downstream failures.

NLP is already production-ready in tools like Jama Connect Advisor™, which uses it to score requirements against INCOSE and EARS rules. ML and LLM capabilities are maturing fast, but they come with data quality and validation constraints that regulated teams need to evaluate carefully before relying on them.

Why AI in Requirements Management Pays Off Early

Most requirements problems start long before coding, and catching them early saves more time than any fix later in the lifecycle. Here’s where teams see the biggest returns:

• Manual effort and documentation time: Some biopharma teams have cut drafting time by up to 70% with generative AI handling data collection and first drafts. For requirements teams, similar savings show up in trace matrix maintenance and review prep.
• Requirements accuracy and consistency: AI-enhanced traceability has reduced review downgrades from 8.7% to 1.6%, and high-confidence trace links increased from 56.4% to 70%. Fewer downgrades means fewer revision cycles on large requirement sets.
• Review cycles and time to market: Writing and testing code accounts for only 25% to 35% of total time from idea to launch, so shortening upstream requirements work has an outsized effect on your schedule.
• Stakeholder alignment: AI can synthesize inputs from stakeholders across different technical backgrounds, flag conflicts between teams, and surface gaps that would otherwise go unnoticed until integration.

Each of these improvements feeds the next. Cleaner requirements lead to fewer test failures, which lead to shorter review cycles, which free up time for the next program.

Challenges and Risks of AI in Requirements Management

AI can do a lot here, but it comes with constraints that matter in safety-critical industries. Three stand out:

• Data quality and training data dependencies: Incomplete training data is a key limiter, with AI-generated requirements omitting core needs when relying on generic datasets. In aviation, emerging guidance calls for data management frameworks addressing bias mitigation and dataset representativeness.
• Over-reliance on automation vs. human judgment: Most AI models remain black boxes, which is a problem in safety-critical industries. LLMs in particular may “generate spurious or hallucinatory material” or fail to comply with established criteria. Human review isn’t optional here. It’s a structural requirement baked into every applicable standard.
• Regulatory and compliance gaps: Current safety standards (ISO 26262, DO-178C, IEC 62304) weren’t written to address non-deterministic AI behavior. Applicants proposing AI software will require FAA involvement, signaling that established means of compliance under DO-178C haven’t caught up yet. Teams adopting AI tools today are operating ahead of finalized regulatory frameworks.

None of these are dealbreakers, but they do mean you should treat AI outputs as inputs to human review rather than finished artifacts.

AI Use Cases in Requirements Management

Here are six specific ways teams are using AI in requirements workflows today, from early-stage elicitation through verification and risk assessment.

Automated Requirements Elicitation and Extraction

NLP can pull requirement candidates out of messy stakeholder notes, meeting transcripts, and regulatory documents. This approach has already been used to accelerate initial requirements work, turning unstructured input into structured, traceable requirement sets. The output still needs human review, but the starting point is much closer to a usable baseline.

Intelligent Document Analysis and Relationship Mapping

Instead of manually cross-referencing hundreds of pages, engineers get an automatically generated relationship map showing how requirements connect to design elements, test cases, and risk items. NLP techniques can now create systems diagrams from documentation, detect ambiguity, link similar documents, and improve quality metrics. For teams managing large document sets, automated mapping cuts the time to answer coverage and completeness questions.

Requirements Quality Scoring and Ambiguity Detection

AI scores each requirement against INCOSE and EARS rules, catching vague terms, passive voice, and missing conditions before anything gets baselined. Without that check, ambiguity survives review and shows up months later when a test engineer can’t write a pass/fail criterion. AI can also scan for near-duplicate or conflicting requirements that human reviewers consistently miss.

AI-Powered Test Case Generation

AI can classify requirements by type, translate them to a logical format, and produce test cases covering nominal, boundary, and failure conditions. In the e-mobility domain, requirements have been used to generate linked test cases without manual authoring. For verification engineers facing hundreds of requirements before a milestone, this turns a multi-week manual effort into hours.

Intelligent Traceability and Impact Analysis

Maintaining end-to-end traceability across requirements, architecture, design, implementation, and test artifacts is one of the most labor-intensive parts of regulated development. AI keeps trace links current by detecting when an upstream change creates a gap or suspect link downstream. When a requirement changes, every affected test case, design element, and risk item gets flagged.

Predictive Risk Identification

AI can surface risk at the requirements phase rather than waiting for testing or a regulatory review. Predictive models flag ambiguities most likely to cause downstream rework, identify missing requirements in high-risk areas, and catch conflicting constraints before they spread. AI can also rank requirements by business value, complexity, and technical risk, giving leads a data-informed view of what to build first and where to cut scope without introducing new risk.

How to Evaluate AI Requirements Management Tools

The real question is whether a tool addresses the failure patterns your team already deals with: ambiguous requirements that survive review, trace links that go stale, and audit pressure when nobody can show what happened and why.

When you’re comparing tools, these three things tell you more than any feature list:

• Integration with existing workflows: Does the tool sync natively with your ALM, issue tracking (Jira, Azure DevOps), PLM systems, and CI/CD pipelines? Requirements changes need to propagate downstream without manual re-entry.
• Traceability and audit trail depth: Bidirectional traceability is a compliance requirement under ISO 26262, DO-178C, and IEC 62304. Look for automated impact analysis, baseline management, and electronic signatures that hold up in a regulatory review.
• Support for your specific standards: Does the tool ship with pre-configured templates aligned to your applicable standards, not generic compliance claims?

If a tool checks all three boxes and also scores requirements quality against INCOSE and EARS, it’s worth a closer look. The fastest way to prove value is to run a quality scoring pilot on a single project. Pick a requirement set that’s about to enter review, score it with the tool, and measure whether the review cycle shortens.

Top AI Requirements Management Tools

The right tool depends on your industry, your existing toolchain, and how much regulatory rigor your traceability needs to support. Here are five tools that come up most often.

1. Jama Connect

Jama Connect is a requirements management and traceability platform built for teams developing complex, regulated products across automotive, aerospace, medical devices, and defense. Jama Connect Advisor scores requirements against INCOSE and EARS standards, generates linked test cases, and flags downstream impacts when upstream items change. Live Traceability keeps the full artifact chain visible across the lifecycle.

Pros:

• AI quality scoring against INCOSE and EARS standards
• Live, bidirectional traceability across the full lifecycle
• Pre-built frameworks for ISO 26262, DO-178C, IEC 62304, and other regulated standards
• Jama Connect Review Center supports structured, auditable review workflows

Cons:

• Designed for complex, regulated programs, so teams without compliance requirements may not need the full depth

Best for: Automotive, aerospace, defense, and medical device teams building safety-critical or compliance-driven products.

2. IBM Engineering Requirements Management DOORS Next

IBM’s cloud-based evolution of the DOORS platform. The Requirements Quality Assistant (RQA) uses Watson AI to score quality and flag ambiguity, passive voice, and missing tolerances during authoring.

Pros:

• Long track record in aerospace and defense
• Watson-powered scoring pre-trained on 10 INCOSE-based quality issues
• Strong configuration management and baselining

Cons:

• Administration and configuration can be complex, especially for occasional users, and teams migrating from DOORS Classic should expect a transition period
• Performance can degrade on large modules with extensive audit history, with some users reporting slow page loads and high server CPU usage during peak activity

Best for: Aerospace and defense programs already invested in IBM engineering tools.

3. Codebeamer (PTC)

A full ALM platform covering requirements, test, and risk management with built-in regulatory templates. PTC acquired Codebeamer in 2022 and has been integrating it into their Windchill PLM ecosystem.

Pros:

• End-to-end ALM with requirements, test, and risk management in one tool
• Strong regulatory templates for automotive (ASPICE), medical devices, and aerospace
• Good Jira and Jenkins integrations for teams running Agile alongside compliance

Cons:

• The full ALM suite can feel heavy for teams that only need requirements management
• Integration with PTC’s Windchill PLM is still maturing, and teams outside the PTC ecosystem may not get the full benefit

Best for: Regulated product development teams that want requirements, test, and risk management consolidated in a single ALM platform.

4. Polarion ALM (Siemens)

Siemens’ ALM platform with requirements management, test management, and change tracking. Polarion integrates tightly with the Siemens ecosystem including Teamcenter PLM.

Pros:

• Unified ALM covering requirements, test, quality, and change management
• Deep integration with Siemens Teamcenter for PLM-connected traceability
• Built-in workflow automation and electronic signatures for regulated industries

Cons:

• Steep learning curve and complex initial setup, especially without existing Siemens infrastructure
• Deployment timelines can be significantly longer than cloud-native alternatives

Best for: Enterprise teams already invested in the Siemens product development ecosystem who need ALM integrated with their PLM.

5. Visure Requirements ALM

An all-in-one ALM platform covering requirements, risk, and test management with a focus on regulated industries. Visure supports ReqIF import/export for data exchange with other requirements tools.

Pros:

• Requirements, risk, and test management in a single platform
• Strong compliance support for DO-178C, ISO 26262, IEC 62304, and other standards
• ReqIF support for requirements data exchange across tools

Cons:

• Smaller user community and partner network compared to IBM, Siemens, or PTC
• Entry-level costs can be higher than lighter-weight alternatives

Best for: Regulated product development teams looking for an all-in-one requirements and compliance platform outside the major PLM vendor ecosystems.

What AI Looks Like Inside an Actual Requirements Workflow

Jama Connect Advisor™ is a good example of what this looks like in practice. When an engineer writes a requirement, Jama Connect Advisor evaluates it against INCOSE and EARS rules, flags vague terms and structural issues, and returns a quality score before the requirement gets saved. The same tool generates test cases from requirements (with steps, linked back to the source), so verification engineers don’t spend weeks drafting them manually. If a requirement changes later, every linked test case gets a suspect flag automatically. Grifols reduced review cycles from three months to fewer than 30 days after bringing Jama Connect Review Center into their workflow.

The underlying idea is that quality checks and traceability should happen inside the authoring workflow, not as a separate exercise before an audit. When those checks run continuously, requirements stay cleaner, trace links stay current, and the team spends less time on rework and more time on the engineering work that moves the product forward.

Getting Started With AI in Requirements Management

If you’re evaluating where AI fits in your requirements workflow, the fastest way to see value is to pilot quality scoring on a single project. Pick a requirement set that’s about to enter review, score it with an AI tool, and measure whether the review cycle shortens and fewer issues come back from the review board.

Jama Connect offers a free 30-day trial that includes Jama Connect Advisor for requirements quality scoring, AI-generated test cases, and Live Traceability across your full artifact chain.

Frequently Asked Questions About AI Requirements Management

Can AI replace human engineers in requirements management?

No. AI catches ambiguous language, missing trace links, and structural issues before they propagate downstream. In regulated environments, human review is a structural requirement. AI reduces the manual burden so engineers can focus on judgment calls that require domain expertise.

What should I look for when evaluating AI requirements management tools?

Three things: native integration with your development environment, support for your specific regulatory standards (not generic compliance claims), and AI scoring grounded in recognized frameworks like INCOSE and EARS.

How does AI improve requirements traceability?

Mostly by keeping trace links current without someone having to manually cross-reference a matrix every time something changes. AI tools maintain those links continuously and flag suspect relationships the moment an upstream requirement is modified, so your team catches gaps in hours instead of discovering them weeks later during a review or audit.

Is AI in requirements management ready for safety-critical industries?

Yes, for quality scoring, traceability, and test case generation. But treat AI outputs as inputs to human review. Regulatory frameworks are still catching up to non-deterministic AI behavior, so use AI for detection and drafting while keeping engineers in the approval loop.

The post AI in Requirements Management: What Works in 2026 appeared first on Jama Software.

Jama Software is always looking for news that would benefit and inform our industry partners. As such, we’ve curated a series of customer and industry spotlight articles that we found insightful. In this blog post, we share an article from AMA, titled “Augmented Intelligence in Medicine” and originally published on October 21, 2025.

Augmented intelligence in medicine

Artificial intelligence vs. augmented intelligence

The AMA House of Delegates uses the term augmented intelligence (AI) as a conceptualization of artificial intelligence that focuses on AI’s assistive role, emphasizing that its design enhances human intelligence rather than replaces it.

AMA policy on AI development, deployment and use

The AMA is committed to ensuring that AI can meet its full potential to advance clinical care and improve clinician well-being. As the number of AI-enabled health care tools continue to grow, it is critical they are designed, developed and deployed in a manner that is ethical, equitable and responsible. The use of AI in health care must be transparent to both physicians and patients.

In addition to medical devices, AI is increasingly used in health care administration or to reduce physician burden, and policy and guidance for both device and non-device use of health care AI is necessary. Recognizing this, the AMA has developed new policy (PDF) that addresses the development, deployment and use of health care AI, with particular emphasis on:

• Health care AI oversight
• When and what to disclose to advance AI transparency
• Generative AI policies and governance
• Physician liability for use of AI-enabled technologies
• AI data privacy and cybersecurity
• Payor use of AI and automated decision-making systems

Physician sentiments on AI

In 2023, the AMA conducted a comprehensive study of over 1,000 physicians’ sentiments towards the use of AI in health care including current use and future motivations for use, key concerns, areas of greatest opportunity and requirements for adoption.​ Given the rapidly evolving AI landscape across health care, the AMA repeated the study in late 2024 (PDF). The objectives of this research remain:

• Capturing the sentiment among practicing physicians regarding the increased usage of AI in health care​
• Evaluating AI use cases based on their familiarity, relevance, and usefulness​
• Identifying key resources and areas of need for physicians to consider implementation of AI tools to their practice

Physicians largely remain enthusiastic about the potential of AI in health care, with 68% seeing at least some advantage to the use of AI in their practice, up from 65% in 2023. We also saw use of AI increase from 38% in 2023 to 66% of physicians reporting they use some type of AI tool in practice in 2024.

However, there are still key concerns as physicians continue to explore how these tools will impact their practices. Implementation guidance and research, including clinical evidence, remain critical to helping physicians adopt AI tools.

Physician sentiments study on AI: AMA’s latest study on physician sentiments around the use of AI in heath care: motivations, opportunities, risks and use cases. Read Now (PDF)

RELATED: AI Accelerates and Enhances Quality of Test Case Generation with Jama Connect Advisor™

AI in medical education

AI is playing an increasingly important role at all stages of the medical education continuum, both as a tool for educators and learners and as a subject of study in and of itself. AI has the potential to transform the educational experience as a part of precision education and transform patient care as a part of precision health. Learn more about how AI can impact medical education.

• In October 2025, AMA launched the Center for Digital Health and AI to put physicians at the center of shaping, guiding and implementing AI tools and other technologies that are transforming medicine.
• AMA welcomes the federal government’s new 2025 action plan on AI and the opportunity to work with the administration to address key areas in shaping AI regulation, policy and implementation. Learn more.
• An AMA issue brief (PDF) provides a brief overview of recent state legislative activity and discusses three key AI policy areas for state legislative/regulatory activity: health plan use of AI, transparency and physician liability.
• To develop actionable guidance for AI in health care, the AMA reviewed literature on the challenges health care AI poses and reflected on existing guidance. These findings are published in a paper in Journal of Medical Systems: Trustworthy Augmented Intelligence in Health Care.

RELATED: The Collapse of Requirements Quality Under System Complexity – How AI Can Help

CPT^® and AI

The current CPT^® code set drives communication across health care by enabling the seamless processing and advanced analytics for medical procedures and services.

AMA offers several resources to provide guidance on the updated CPT^® code set for classifying various AI applications as well as advisory expertise through the Digital Medicine Payment Advisory Group (DMPAG). DMPAG identifies barriers to digital medicine adoption and proposes comprehensive solutions on coding, payment, coverage and more. Stay up-to-date on the criteria for CPT^® codes, access applications and read frequently asked questions.

The post Augmented Intelligence in Medicine appeared first on Jama Software.

This blog recaps a preview of this webinar, watch the entire thing here “From Requirements to Regulatory: How AI Is Transforming Submission Readiness”

Stop Scrambling for Submissions. Build Readiness Into Your Process With AI.

Regulatory submissions often become a stressful, last-minute rush, increasing risk, rework, and frustration. But what if you could embed submission readiness into your process from the start? Artificial Intelligence (AI) is making this a reality by connecting requirements, regulatory guidance, and ongoing monitoring seamlessly throughout the product lifecycle.

In this webinar, Adam Smith, CEO of Agent Astro, and Tom Rish, Senior Business Development Manager at Jama Software, share practical insights into how AI helps teams stay submission-ready, reduce delays, and maintain compliance without slowing development.

Key Takeaways:

• The benefits of integrating submission readiness early to save time and reduce stress.
• How AI can track requirements and ensure traceability between design decisions and regulatory guidelines.
• Tips for maintaining compliance as new regulations emerge and post-market data is collected.
• A look at how AI will continue to shape regulatory processes in the future.

WEBINAR PREVIEW BELOW – WATCH ENTIRE WEBINAR HERE

TRANSCRIPT PREVIEW

From Requirements to Regulatory: How AI Is Transforming Submission Readiness

Tom Rish: Thank you to everyone for being here today. We have a very exciting webinar about AI, a hot topic, of course, as always, and so I’m excited to dive into it. Before we do, I just want to talk very briefly about Jama Software and what we do. I know some of you have watched previous webinars, and you know all about this, but I want to give a high-level overview and talk just a little bit about how we are looking to incorporate AI to make your life easier when it comes to requirements management. So first, Jama Connect ^®. As you all know, when it comes to launching a product, you have to keep track of all your requirements, all of your risk items, all of your testing, and everything like that. It can be a lot of work, especially on spreadsheets or disjointed systems, whatever it is you use.

And at Jama Software, what we’re trying to do is make it simple for you. We want you to focus on designing. We want you to focus on testing. We want you to focus on important things like the safety of the patient and not worry as much about paperwork and organizing everything. A lot of times, as you know, that’s done at the end, and it’s a checkbox activity. But we have a system, as you can see there on the left. I know many of you are used to a lot of documentation and everything. We want to bring that into a very organized V model that you’ve all seen there. Start with user needs. Enter those right into the system, build as you go. We can connect all of the systems you use, whether it’s software products, and you’re using a lot of things like Jira, GitHub, things like that, all your test systems, but we want to keep things organized.

RELATED: Buyer’s Guide: How to Select the Right Requirements Management and Traceability Solution

Rish: What’s cool about Jama Connect is that we work with all industries, but we have frameworks specifically for medical devices. So out of the box, we’re able to build a framework where you can match it to your processes to track your user needs, design controls, risk management, and all of your tests. We have real-time collaboration so that you can do all of your reviews and comments in the software, create libraries, and release things. And finally, we have the AI guidance that I’m here to talk about today.

A couple of things here on this slide. This is mostly focused on requirements management. One of them is there today and available for use. Some of our customers are using it, and we’ve gotten some good feedback. Some of these here are things that are coming in the future. First thing that we have here today, though, is a scoring system. So when you enter your requirements into Jama Connect, we have AI that scans through INCOSE and EARS guidance and tells you how well this requirement is written. So it gives you a scoring system to tell you, “Hey, this one looks pretty good,” or, “This one doesn’t, and here’s why this is the rule or the guidance that it doesn’t quite meet.” So that’s ready to use today. I’ve talked to a few customers already who have said how helpful it has been for downstream operations like testing to create better testing and things like that.

We’re also working on some things where we will help rewrite requirements if needed. So not only does it give you scores, but help you rewrite them so that they can match the guidances better. So if you give it an initial draft of a requirement, we’ll go through, we’ll score it, but we’ll also give you some recommendations for changing it.

I think ideally everyone’s probably wondering how you just create them for us. So we are looking into some ways that we can enter some project inputs into the software, and then it will give you some requirements for you. So that will be in the future, along with PDF parsing. A lot of you come with existing documentation already. You might have requirements documents, software specification documents, things like that. We’re working on some AI features that will take those and create requirements automatically for you in the structure that they are.

RELATED: Empowering Complex Development with Responsible AI

Rish: A couple of other things. One thing that is new now, again, is test case generation. When you have your requirements in there, what we want to do is help you create good testing and guidance for creating the right acceptance criteria and things like that for your testing. Also, looking at an AI assistant, I think everyone is used to AI assistance these days, but a more conversational workflow where you can enter information into the software, and we’ll give you some guidance and feedback on that. Also, looking into ways that we can take your requirements and give you tips on how to link them together better, create better relationships, and finally help with reviews to detect areas that maybe are high risk.

I think later on, what we’re going to talk about is how the FDA and other regulatory bodies are starting to incorporate AI. So what we want to do is help you get it right up front so that when it’s sent over there, you feel good about everything. So that’s a little bit about Jama and how we’re using AI today. Now for the main event, I’m excited to pass it over to Adam. I met Adam at the MedTech conference in San Diego. And when I went up to his booth, I was instantly impressed. I think as a product development engineer, I spent a lot of time searching through the FDA databases.

And there are a lot of them, as I’m sure you all know, and there is excellent information in those databases. The challenging part is that it’s hard to go to each one every time and find what you need. The interfaces are a little outdated at times as well. You can find everything, but it’s just not easy. And what I always thought is, why can’t anybody scrape this information or pull this information and use it in a better format and make our lives easier? And that’s exactly what Adam and his team are doing. And so I’m excited to hand it over to him, and he will tell you more about Agent Astro and give some practical tips about how to better use AI throughout your process.

TO WATCH THIS ENTIRE WEBINAR, VISIT:
From Requirements to Regulatory: How AI Is Transforming Submission Readiness

The post [Webinar Recap] From Requirements to Regulatory: How AI Is Transforming Submission Readiness appeared first on Jama Software.

Navigating FDA AI Guidance for Medical Devices: A Practical Guide

For medical device professionals, the integration of Artificial Intelligence (AI) and Machine Learning (ML) represents a monumental leap forward in innovation. However, this progress comes with significant regulatory hurdles. As AI algorithms evolve, so do the rules that govern them, leaving many development, quality, and regulatory teams struggling to keep pace. Failing to understand and adapt to the latest FDA AI guidance can lead to submission delays, compliance issues, and costly rework.

This guide delivers a practical overview of the evolving FDA regulatory framework for AI and ML-based medical devices, drawing on both recent draft guidance and the agency’s longer-term action plans. We highlight essential concepts including the Predetermined Change Control Plan (PCCP), Good Machine Learning Practices (GMLP), and Real-World Performance (RWP) monitoring and show how these shape the compliance landscape for manufacturers.

TL;DR: The FDA is moving toward a holistic Total Product Lifecycle (TPLC) regulatory approach for AI/ML-enabled medical devices, emphasizing continuous monitoring, clear GMLP, and mechanisms for pre-planned algorithm updates. Robust, traceable documentation, and proactive lifecycle risk management are now essential for compliance and product success.

The FDA’s Evolving AI/ML Regulatory Framework

The FDA has signaled its commitment to adapting device oversight in response to rapid advances in AI/ML. Traditionally, regulatory submissions were point-in-time events. Now, regulators recognize that adaptive, learning systems require ongoing oversight, especially as software “learns” from real-world experience.

Key foundational documents illustrate this evolution:

• FDA’s 2021 AI/ML-Based Software as a Medical Device (SaMD) Action Plan: This action plan lays out five pillars to modernize oversight including development of a tailored regulatory framework, advancement of GMLP, fostering transparency with users, promoting methodologies for bias/robustness, and supporting real-world performance pilots.
• Artificial Intelligence-Enabled Device Software Functions: Lifecycle Management and Marketing Submission Recommendations (Draft Guidance, 2025): This draft guidance details expectations for managing AI within medical devices throughout the entire product lifecycle, including design, labeling, bias mitigation, cybersecurity, postmarket surveillance, and the importance of the Predetermined Change Control Plan.
• Clinical Decision Support Software Guidance (2026): Clarifies FDA’s criteria for Clinical Decision Support (CDS) software functions, offering practical examples to distinguish between Non-Device CDS such as software functions excluded from device regulation and those that remain under device oversight.
• FDA AI/ML-Enabled Medical Devices List: Provides a current catalog of FDA-authorized devices using AI/ML technologies, helping manufacturers benchmark their projects and understand regulatory precedent.
• General Wellness: Policy for Low-Risk Devices: Clarifies what qualifies as a low-risk wellness device or feature and what falls outside medical device oversight.

In summary: The FDA’s approach now encompasses both initial submissions and ongoing, risk-based management, aligning regulatory expectations with the unique characteristics of AI/ML-driven technologies.

RELATED: The Evolution of FDA Design Controls (21 CFR 820.30) and How Jama Software Supports Compliance

Core Concepts in the FDA’s AI/ML Oversight

1. Predetermined Change Control Plans (PCCP)

Introduced in both the 2021 action plan and expanded in the draft 2025 guidance, a PCCP enables manufacturers to define anticipated modifications to an AI/ML algorithm upfront. The plan specifies “what” may be changed (pre-specifications) and “how” changes are managed (an algorithm change protocol). This approach recognizes the evolving nature of AI/ML models, especially those learning from real-world use.

2. Good Machine Learning Practices (GMLP)

The FDA calls for GMLP, which are best practices covering data management, training procedures, documentation, interpretability, and bias mitigation, all aligned with consensus standards. GMLP underpins both product quality and regulator confidence, reducing the risk of unexpected outcomes or patient harm (See Action Plan Pillar 2).

3. Transparency and User Trust

Both guidance documents emphasize transparency for end users including clinicians, patients, and caregivers. Clear labeling, robust documentation, and transparency about model logic, data sources, and limitations are expected to build trust in AI/ML-powered devices.

4. Real-World Performance (RWP) Monitoring

Unlike static software devices, AI/ML-based products must demonstrate ongoing safety and efficacy. The FDA encourages collection and review of real-world data as part of postmarket surveillance. Manufacturers should implement plans for ongoing performance monitoring by adapting both processes and documentation to ensure device quality over time.

5. Bias Mitigation and Robustness

AI/ML algorithms can inadvertently encode biases from historical datasets. The FDA expects proactive identification and management of bias through diverse, representative training data, ongoing performance validation, and transparent reporting on limitations and subgroup analysis.

RELATED: Buyer’s Guide: Selecting a Requirements Management and Traceability Solution for Medical Device & Life Sciences

Practical Strategies for AI/ML Medical Device Teams

Step 1: Build a TPLC-Ready Development Program

Map your design and postmarket processes to the FDA’s TPLC vision. Reference both the 2025 draft guidance and the 2021 action plan to ensure full coverage.

Step 2: Document Everything—and Connect the Dots

Your design history, risk management, GMLP adherence, model versions, data sets, and algorithm updates should all be auditable and linked. Use digital solutions for traceability and compliance, making audit preparation seamless.

Step 3: Prepare and Maintain a PCCP

If your product uses adaptive algorithms, develop a comprehensive Predetermined Change Control Plan. Detail the types of future modifications, associated risk controls, and your process for validating postmarket changes.

Step 4: Embrace Ongoing RWP Monitoring

Postmarket surveillance now means real-world performance tracking including collecting user feedback, monitoring for data drift, bias, and managing field updates in a proactive, traceable way.

Step 5: Differentiate Wellness from Medical Claims

Consult the Wellness Policy to determine if any features of your device are exempt from device regulation and document your rationale.

Frequently Asked Questions

Q: What’s the difference between Software as a Medical Device (SaMD) and AI in Medical Devices (AiMD)?
A: SaMD refers to software that is itself a medical device. AiMD is software that is integrated into a physical device. Both fall under the FDA’s AI/ML regulatory frameworks.

Q: Is a PCCP mandatory for all AI-enabled devices?
A: PCCPs are expected for devices with adaptive/evolving algorithms. Rigid, non-learning AI products may not need a PCCP, but processes for documenting and justifying updates are still required (draft guidance, 2025).

Q: How should we implement GMLP?
A: Follow best practices outlined by the FDA and consensus standards. Ensure your team manages data, training processes, versioning, and labeling in a repeatable, controlled, and demonstrable manner.

Master the Complexity of AI Medical Device Development

The regulatory landscape for AI medical devices is complex, but it shouldn’t stifle innovation. By adopting an integrated approach with a live digital thread, you can manage the intricate web of requirements, risks, and data that define modern device development. This not only prepares you to pass audits with confidence but also empowers your teams to build safer, more effective products faster.

Jama Connect^®, enhanced with AI-powered features in Jama Connect Advisor™, provides the end-to-end traceability needed to manage the development of complex AI-enabled systems. Streamline your documentation, automate traceability, and ensure your team is always audit-ready.

Ready to see how you can transform your product development process? Schedule a personalized demo to learn more about Jama Connect.

Note: This article was drafted with the aid of AI. Additional content, edits for accuracy, and industry expertise by Tom Rish.

The post Navigating FDA AI Guidance for Medical Devices: A Practical Guide appeared first on Jama Software.

This blog recaps a section of our webinar, to watch the entire presentation, visit The Collapse of Requirements Quality Under System Complexity – How AI Can Help​

Transforming Requirements Engineering with AI to Enhance Clarity, Consistency, and Scalability

As systems grow more complex, traditional processes struggle to keep up, ultimately impacting requirements quality.  AI can assist in processing the sheer volume of data, enhancing clarity, consistency, and scalability across workflows.

Join Katie Huckett, Product Line Manager for Advisor/AI at Jama Software, for an exclusive webinar exploring how AI is becoming an essential cognitive amplifier in requirements engineering. Discover how AI is redefining the way teams detect ambiguity, surface hidden conflicts, and maintain alignment at scale.

What You’ll Learn:

• Understand why requirements quality is declining under modern system complexity.
• Learn the hidden costs of poor requirements and why traditional practices fall short.
• Discover how AI amplifies cognitive processing and improves requirements quality.
• Explore practical steps for adopting AI in your engineering workflows.
• Gain insights into the future of requirements engineering with AI.

The video below is a preview of this webinar, click HERE to watch it in its entirety

WEBINAR TRANSCRIPT PREVIEW

The Collapse of Requirements Quality Under System Complexity – How AI Can Help​

Katie Huckett: Welcome, and thanks for joining. Today we’re going to talk about something many engineering organizations are experiencing, but rarely say out loud. Requirements quality is collapsing under the weight of modern system complexity. This session isn’t about tools, features, or automation for automation’s sake. It’s about why this problem exists, why traditional fixes are no longer sufficient, and why AI is becoming a necessity rather than a nice to have in requirements of engineering.

My name is Katie, and I lead product strategy focused on AI-driven capabilities and requirements management. I spend most of my time working with engineering teams in highly regulated complex industries, aerospace and defense, automotive, medical devices, and other systems where requirements quality is not optional. What I’m sharing today is based on what those teams are actually struggling with in practice, not theory.

Here’s how we’ll spend our time together. We’ll start looking at why requirements quality is breaking down despite increased process maturity. We’ll talk about the hidden costs of complexity and why traditional approaches no longer scale. Then we’ll look at how AI changes what’s possible, not as a replacement for engineers, but as a cognitive amplifier. And finally, we’ll discuss what this shift means for engineering organizations moving forward. We’ll have a brief Q&A portion before we conclude today. Let’s dive in.

Here’s the paradox we’re living in. Requirements practices are more mature than they’ve ever been. Teams have invested heavily in process, tooling, standards, and governance, and yet many organizations are seeing more rework, more late stage surprises, and more friction between teams than before. What’s important here is that this isn’t happening because teams stopped caring about quality. It’s happening because the nature of the systems we’re building has changed faster than the way we manage requirements. In other words, the rules of the game changed, but most practices did not.

Modern products are no longer confined to a single domain. A single system now routinely spans software behavior, physical components, data flows, safety constraints, regulatory requirements, and operational considerations. All of these elements evolve together, often on different timelines and often with different teams responsible for each part. As systems scale and change in parallel, the number of relationships between requirements increases dramatically, not linearly. And yet, many traditional approaches still assume that these relationships can be reasoned through manually during periodic reviews or checkpoints. The challenge isn’t capability or commitment. It’s that the structure of the work itself has fundamentally changed.

RELATED: Jama Connect Advisor™ Datasheet

Huckett: Before we go further, I want to ground this discussion in your experience. We’re going to launch a poll. Please take a moment to answer honestly. What is the biggest contributor to requirements quality issues in your organization?

Looks like we have the results in. In nearly every organization I work with, the answer is rarely just one of these. These challenges stack on top of each other, and that compounding effect is exactly what overwhelms traditional requirements practice.

Traditional requirements practices were built for a world where change was slower, and systems were more predictable. Reviews happened at defined milestones. Documents were relatively stable. Dependencies were fewer and easier to reason about. Today, however, requirements are changing continuously, often across teams working in parallel. When you apply periodic document-centric review models to this environment, gaps are almost inevitable. The process itself isn’t wrong. It’s just being asked to operate outside the conditions it was designed for.

It’s important to say this clearly. This is not a lack of skill problem. It’s not a lack of effort problem. It’s not a lack of accountability problem. It’s a structural mismatch between human cognitive limits and the complexity of modern systems.

One of the most dangerous things about requirements quality issues is that they rarely fail loudly. A single ambiguous requirement doesn’t stop a project. It quietly creates multiple interpretations. Those interpretations propagate into design decisions, test cases, and validation activities. By the time the issue is discovered, multiple teams have already invested time and effort based on different assumptions. And at that point, the cost isn’t just fixing the requirement. It’s undoing everything that was built on top of it.

RELATED: Buyer’s Guide: How to Select the Right Requirements Management and Traceability Solution

Huckett: Let’s do another quick poll. Where do requirements quality issues most often surface too late in your lifecycle?

Some interesting results here. Wherever this shows up in your lifecycle, the pattern is consistent. Humans don’t see the issue until it’s already costly. That’s not a vigilance problem, that’s a visibility problem. When quality issues surface, the instinctive response is to add more safeguards. That means more reviews, more sign-offs, more documentation. The problem is that these measures increase effort without increasing visibility. Teams end up spending more time checking artifacts, but not necessarily improving quality or alignment. In highly complex systems, quality doesn’t improve by adding friction. It improves by improving signal.

This is where AI fundamentally changes the equation. AI doesn’t get tired. It doesn’t lose focus. It doesn’t skip over sections because a document is long or familiar. It can continuously scan requirements, compare them, and look for patterns or anomalies across the entire system. That doesn’t replace human expertise. It supports it by ensuring that engineers are spending their time where judgment actually matters. In that sense, AI becomes part of the engineering infrastructure rather than a separate tool.

TO WATCH THE ENTIRE WEBINAR, VISIT:
The Collapse of Requirements Quality Under System Complexity – How AI Can Help

The post [Webinar Recap] The Collapse of Requirements Quality Under System Complexity – How AI Can Help​ appeared first on Jama Software.