When AI Signals Risk

This interview is part of our extended Q&A series exploring AI and food safety.

As AI systems become more capable of identifying early warning signals, they also raise new questions about responsibility. If a system flags a potential issue, companies may be expected to investigate and respond, even if no contamination has occurred. In this Q&A, Shawn Stevens, founder of Food Industry Counsel LLC, explains how AI could shift legal risk and why responses to these signals matter.

How should companies think about relying on AI in food safety systems?

If AI generates, let’s say, a food safety program for me and facially it looks like it is sufficient—it identifies hazards, proposes various controls, and tells us how to monitor and verify that those controls are effective—and we rely upon that, well, if it’s wrong and we produce unsafe product or people get sick, we’re still going to be the ones liable.

That’s where the reliance part comes in. Trust is really another form of reliance. If you’re relying on AI to perform a function and you’re assuming that function has been performed correctly, then the legal responsibility still stays with you.

Are there lower-risk ways to use AI in food safety?

Yes. On the positive side, I remember hearing food safety professionals from sophisticated organizations talk a few years ago about using AI to monitor employee conduct on the processing floor or in restaurants. That could include things like whether an employee is washing their hands at the appropriate intervals or after certain tasks, whether they’re washing correctly, whether tables are actually being sanitized after customers leave, or whether a sanitation crew is cleaning to the standard created by the sanitation program.

If we look at AI through that lens—as an added layer of monitoring—the legal risk is very low. We’re taking a program that already exists and employees who are already doing what they’re doing, and in belt-and-suspenders fashion, we’re layering in additional monitoring with some alarms in case something isn’t quite right.

I would be very comfortable standing in front of a jury and saying: This company had the option not to use AI, but it decided it wanted to get better and wanted to make sure that if mistakes were being made or shortcuts were happening, it could identify those and take appropriate action. That kind of use is easily defensible. But if we’re leaning on AI to do things that humans, food scientists, or food safety and quality professionals have historically done—and we’re trying to remove the human from that space—that’s where I think we’re fraught with extreme legal liability.

If AI tells you there’s a risk and you don’t act, that’s where exposure comes in.

Shawn Stevens

Food Industry Counsel LLC

Can AI increase legal exposure in the event of contamination?

Yes, I think it can. I’ve spent a lot of time over the years talking about potential Department of Justice criminal exposure for food companies when they produce products that make people sick.

Take a deli product with intermittent Listeria positives in the environment, but no finished-product positives, so the company keeps shipping. Over time, there are repeated positives, but everything appears fine until eventually people get sick. Then the U.S. Food and Drug Administration (FDA) comes in, the Department of Justice (DOJ) comes in, and they assess whether there’s a basis for criminal sanctions. Under the Park Doctrine, [which allows the FDA to hold corporate executives criminally liable for misdemeanor violations of the Federal Food, Drug, and Cosmetic Act without proof of personal knowledge, intent, or direct participation in the violation], there are basically three questions: Were you aware of a circumstance that could lead to contamination? Were you in a position to eliminate that condition? And did you fail to do so?

That’s the typical pre-AI scenario. Now layer in AI. If AI identifies trends that aren’t visible to the human eye and [indicates], based on traffic patterns and conditions in the facility, there is a heightened risk for product contamination, does that trigger awareness? I think it might.

Second, is the company in a position to act in response to the AI information, even if there’s no finished-product contamination yet? And third, if the company fails to act, then you could be looking at not only civil exposure but potentially DOJ exposure as well.

So yes, I do believe there is heightened legal risk when we start trusting or relying on AI in the performance of food safety duties.

Could AI also reduce liability if companies use it properly within their food safety systems?

I think the answer turns on how we interact with AI. This is the same problem we’ve had in the food industry forever: We’re really good at collecting data, but then it sits in spreadsheets or files and nobody analyzes it. AI creates the same issue. If it starts telling us things and we choose not to believe it—and it later proves to be right—we could absolutely be scrutinized.

If a client came to me and said, “Our AI system is telling us there may be something we need to look at more closely,” my advice would be to conduct a full food safety assessment—analyze the data, perform root-cause analysis, and take corrective action if needed. If we document that process—what we saw, what we did, and why—we’re in a strong position. Even if we’re ultimately wrong, regulators and juries are more likely to accept that we acted responsibly and in good faith.

The key is not to be complacent or dismissive. We need to take the signal seriously, investigate it, and document our decisions.

What concerns you most about how companies might adopt AI?

I would advocate that we, as an industry, should always be willing to embrace tools that help us do a better job. Where I’m concerned is when companies start using AI primarily as a cost-cutting tool. If the mindset becomes, “How can we reduce QA staff or save money by relying on this technology?” rather than improving food safety, that’s where we get into dangerous territory. If companies are rolling out AI and essentially rolling the dice, that’s where risk increases significantly.

Five years from now, what role do you expect AI to realistically play inside food safety programs?

I think we’ll see certain applications of AI become accepted and standardized across the industry. The technology will mature, and we’ll gain confidence in specific use cases. Some of those may become best practices, similar to how environmental monitoring programs are widely accepted today. At a minimum, AI will be valuable for analyzing the large amounts of data we already collect and helping us interpret it more effectively.

Beyond that, I think we’ll see broader applications in supplier management, monitoring records, and operational data analysis. AI can also be layered onto existing technologies—like sensors that detect equipment issues—and help identify early warning signs before problems occur.

There will still be challenges, especially in more subjective areas like environmental control, where experience and instinct matter. But overall, I expect AI to play a significant role in strengthening food safety systems and helping companies produce safer, higher-quality products.

This interview has been edited for clarity and brevity.