^{Published on 17.3.2026}

As we speak, artificial intelligence is hard at work terraforming the medical device landscape. While its potential is well known and much discussed, as with any transformative technology, AI introduces new complexities that manufacturers must address early and thoroughly. Luckily, it’s nothing you can’t get through with expert guidance.

To shed light on the key challenges and regulatory expectations, Innokas QA/RA Specialist Sandra Hänninen shared her insights on what it really takes to design, validate, and bring AI driven medical devices to market in an interview. Sandra has also been active in panel discussions on this topic.  

‍

Understanding the unique risks of AI-driven medical devices

While AI brings capabilities beyond traditional algorithms, it also introduces entirely new categories of risk. According to Sandra, one major concern is how AI can impact patients’ fundamental rights, especially when performance varies across groups or conditions. Let's name some of those risks here.

1. Bias in training data

AI models are only as good as the data used to train them.
If the dataset is incomplete or unrepresentative, the device may perform inconsistently across patient groups, for example, different ages, ethnicities, or clinical contexts. This can (and will) compromise both safety and equity.

2. Data drift over time

Real world data changes.
Patient demographics shift, new clinical patterns emerge, and operational environments evolve. Without proper monitoring, an AI model’s performance may degrade after deployment, leading to less accurate and overall unreliable outputs over time.

3. Lack of transparency

Medical professionals and regulators must be able to understand how an AI-driven system arrives at its conclusions.
Opaque “black box” models make it harder to validate safety and effectiveness and make it more difficult for clinicians to use outputs responsibly.

4. Cybersecurity vulnerabilities

AI systems often rely on large datasets and networked infrastructures, which in turn makes them attractive targets for cyberattacks.
Breaches can (and again, will) affect both patient data confidentiality and device functionality.

To address these risks, manufacturers must implement a robust, AI-aware risk management process that addresses these unique challenges throughout the product lifecycle.

‍

What regulators expect from AI-driven medical devices

Notified bodies are increasingly familiar with AI-driven technologies, but they also expect more rigor and clarity than ever before. Sandra highlights several areas where manufacturers should be ready to provide detailed documentation.

1. Clear description of the model

Regulators want a precise, technical explanation of:

• The chosen algorithm or model

• All input and output variables

• Performance metrics related to the intended use, such as accuracy, sensitivity, and specificity

2. Evidence of appropriate training data

Manufacturers must demonstrate:

• How the training data was collected

• That the data meaningfully represents the target population

• That clinical and operational conditions match expected real-world use

3. AI-specific risk management

Traditional risk management is not enough when it comes to devices that incorporate AI. Regulators expect documentation on risks unique to AI, including:

• Potential sources of bias

• Data drift and long-term performance risks

• Cybersecurity threats

• Mitigation plans and monitoring strategies

4. Thorough documentation

Every design choice, dataset, performance benchmark, and mitigation action must be recorded. Regulators rely on this documentation to assess the device’s safety and effectiveness, as well as the manufacturer’s overall maturity in managing AI-based technologies.

‍

Responsibly towards evolved healthcare practices

AI offers powerful opportunities to enhance medical devices, but this can happen only when innovation is paired with diligent preparation. Understanding the regulatory landscape, anticipating AI-specific risks, and maintaining transparent documentation are essential steps toward successful approval and safe patient use. This early groundwork leads to better, more trustworthy products that clinicians and patients can rely on.