Joyson Safety Systems on AI in Manufacturing and Cybersecurity

Chad Watt:

Welcome to the Infosys Knowledge Institute podcast, where business leaders share what they've learned on their technology journey. I'm Chad Watt, Infosys Knowledge Institute researcher and writer. Today, I am speaking with Maria Haight, Global Information Security Officer at Joyson Safety, a leading maker of seatbelts, airbags, and related automotive safety products, and Rakesh Babu Gollapalli, Senior Vice President in the Manufacturing Vertical here at Infosys. Welcome to you both.

Maria Haight:

Thank you.

Rakesh Babu Gollapalli:

Thank you.

Chad Watt:

Maria, what is Joyson's posture toward artificial intelligence?

Maria Haight:

So our leadership, starting at the top with our CEO, Philip Shan, is excited about the potential to think differently and find efficiencies with AI tools in the workplace. Our AI motto is better output, better product, better decisions. But we're definitely putting on our security hats on too to make sure our proprietary or sensitive data stays secure. This generated actually the origination of an AI governance board earlier this year to have a process to bring the desired use of our AI tools in the workplace with proper review across multiple domains such as security, compliance, engineering, legal, technology and product development, marketing and manufacturing. And then we get to agree together if we're going to move forward with our proof of concept or even a limited engagement for our intended users until we're comfortable with the accuracy and usefulness of the tool. You're finding there may not be a one size fits all AI tool that works across all of our teams.

Chad Watt:

Rakesh, where are you seeing AI put to work in manufacturing?

Rakesh Babu Gollapalli:

Chad, when I talk about AI in manufacturing, the best way to explain it would be to look at it in three different areas. Manufacturers are using AI in their end products. They are using AI in the actual manufacturing process itself, and then AI in everything supporting manufacturing. So let me drive this by giving you some examples. Let's look at AI in end products. On the automotive side, look at driver assistance systems that helps lane keeping, adaptive cruise controls, collision avoidance, EV battery management, the whole autonomous driving systems. So all these are AI in end products. Similarly, if you look at on the home front, you have Robo Vacuums, you have Smart Thermostats, and actually great advances in healthcare manufacturing. You look at wearable ECGs, smart hearing aids. I don't know if you had a chance to look at Oral-B's Genius X. It's an AI-driven brush that understands the patterns and the missed areas. So it recommends the changes you need to do in your next brushing cycle. Similarly, AI in washers and dryers. AI helps detect fabric types, load sizes, dirtiness levels, and optimizes the wash cycles. So no more guessing there. Again, AI in automotive accessories. You have dash cams that are able to detect dangerous behavior and alert drivers. So there's a lot going on AI in the end product.

Now, when you look at AI in the manufacturing processes themselves, while it is pervasive across the board, there are four key processes I want to highlight. One is product engineering, second is sales, third is quality, and the fourth is predictive or preventive maintenance. So let's look at product engineering. It's a knowledge intensive process, where we have a lot of input data that is available from IoT devices from multiple sources. And GenAI is being leveraged here to accelerate time to market and improve productivity overall in the product engineering side.

On the sales front, break it up into two pieces. One is B2C sales and another is B2B sales. In the B2C sales, AI is helping actually contextualize the responses to the customers. It's helping improve cross-sell and upsell there. In the B2B world, AI is helping how you respond to RFPs by introducing more standardized parts, standardized platforms, and standardized solutions. Again, improving win rates and reducing the cost to respond.

On the quality, probably this is where we are seeing significant AI investment. For example, manufacturers are using computer vision to reduce the cost of poor quality. And a lot of OEMs are actually using this in their welding processes, in castings, in forging processes to detect and eliminate errors right upfront. Then of course, in the predictive maintenance where actually AI is helping OEMs and other manufacturers moving to need-based maintenance rather than scheduled maintenance. And that's again, big savings in terms of downtime.

Lastly, we talked about AI for supporting manufacturing. AI is being leveraged for energy optimization. Some OEMs are actually using AI to fine tune their compressed air systems. This in turn reduces the electricity usage. AI is optimizing the energy leveling that happens to different machines and different plants. And probably the most important and interesting thing is AI being leveraged in supply chain scheduling so that manufacturing gets material just in time, which is what everybody wants from an inventory optimization perspective. So different ways AI is being pervasive in manufacturing.

Chad Watt:

Now, Maria, as the global information security officer for a manufacturing company, you manage the governance of AI. How are you tackling this?

Maria Haight:

So being part of our AI governance board, I'm specifically reviewing third party AI tools from a data and technology security perspective. I want to know things like what data are we looking to utilize? Can our data stay within our environment? Any data that we put in an AI prompt, is it being stored? And if so, it be purged? But I do wonder to myself, how many other companies are taking the time to do the same research and spending resources doing so, right? When is the SOC 2 for AI tools going to come out so we can all leverage and work with a similar methodology, right? Our vetting process at Joyson, I think, is well thought through. But what about all the existing tools in our environment that all of a sudden now have an AI component available? So we are looking at things such as CASB tools that can be used to auto block generative AI tools that are not company approved and DLP rules for monitoring what content is being placed into AI prompts. So those are the things we're looking at currently for safeguards.

Chad Watt:

So Rakesh, AI is changing quickly, and capabilities are growing on a daily basis. Can you identify some constants in AI, some things that are not changing with AI?

Rakesh Babu Gollapalli:

I couldn't agree better on that. AI is changing quickly and it's changing by the day. But there are a few constants and again, I'll try to list the top four or five that we see. The first one is of course, data challenge. AI lives off of data. So getting data ready for AI is very important for everyone, but more so for manufacturers as they deal with diverse data sets. It's ranging from design documents, design manuals, IoT data, and then of course the merit systems that they have, which is the OT systems, the CART systems, the custom systems, so on and so forth. So data challenge is a constant.

Second, as Maria also highlighted, data security challenge. Again, this is a very big challenge for manufacturers in two ways. One is manufacturing today is transforming from mechanical manufacturing to more software-defined manufacturing, right? So it means everything is becoming software defined. So, the security challenge on the end product is paramount. Second is, of course, manufacturers are using AI for their internal processes. Again, what this means is if an OEM is sending you a drawing, it's very important to keep that secure because you don't want your engineers to be able to access design from one OEM while designing the product for the other OEM. So very, very important data security challenge. In fact, in 2024 alone there were close to about 100 ransomware attacks and more than 200 data breaches just for automotives and the mobility ecosystem. So this threat and this challenge is real and it's a constant for AI world.

The third one is the coexistence of automation and AI. AI is not going to be the solution for everything. So you have to be very careful on where we use AI, where we use automation and where we use a combination of them.

Now last but not the least, it's all about the models. We hear a large language model, a small language model, a general purpose models, and so on and so forth. And this is a reality. In fact, the general purpose models are becoming very powerful in a lot of big investments going on. And with every new release, you have greater features in there. In fact, GPT 5.0 got released recently, as you are aware. So very powerful. But that said, every manufacturer has certain nuances, including your individual companies. They have certain nuances. And sometimes these general purpose large language models do not understand those nuances. This is when we start thinking about purpose-built models, or sometimes also called small language models. So these are the four key constants. I'm sure there are several more, but I think these are top of mind right now.

Chad Watt:

Now, Maria, let me ask you to focus in. We've talked about AI across all these different uses. Specifically in the context of cybersecurity, how do you put AI to work?

Maria Haight:

That's great, but I have to brag about our team just a little bit, Chad. So we are a global company, right, in 60 locations, and there's many departments, which is so exciting for us, that really visualize that they would like to use AI to assist their teams. So some of the experiments we've been seeing across the company are things like patent specification drafting, flow chart conversion, research and analysis of literature and statutes. So our legal team will come forward for things like that. Visual and graphic generation, presentations, of course, and communication support, but repetitive engineering and test case creation. So a lot of IT teams and technology teams are really seeing tasks that were taking days go down to hours, tasks that taking hours go down to minutes. But on my side, then the cybersecurity, I'm personally looking at agentic AI solutions to mature my SOC team. How do I reduce the noise of all those alerts, specifically false positives? How can I help them with reducing repetitive tasks? And that's where the AI agents for our SOC group is really starting to mature and become exciting to us.

Chad Watt:

Rakesh, you have some thoughts. Can I ask you to reflect on AI for cybersecurity?

Rakesh Babu Gollapalli:

Absolutely. I think Maria, you hit everything on there, you know, and I'd just like to take it a bit broader here. We talked about how everything in manufacturing is becoming software defined. And what that means is the value of software in a manufactured product is becoming more valuable than the hardware that we are manufacturing actually. So it's software defined vehicles, software defined networks, software defined industrial systems. What this means is that the products are highly connected. They are constantly updated and having very complex attack surfaces. So that's the challenge for a CISO like Maria, right? I mean, that attack surface is just growing multi-fold. So I think this is where AI can really help on the cybersecurity front.

In addition to this, manufacturers, as we just spoke about, have an extensive application landscape. This ranges from COTS product on one end to custom applications and antiquated shop floor systems, so on and so forth. So the software defined product itself, then the complex processes, these are all big challenges for cybersecurity. And as Maria mentioned, AI is going to be the best friend as far as I am concerned. Because it is able to detect patterns that humans just find almost difficult considering the quantum of data that they have to sift through.

Maria talked about the SOC or the Security Operating Center. There are various things they can do by leveraging AI, right? It's predictive threat intelligence. It's threat anomaly detection. And on the OT side, is there any network pattern detection they can do? Insider threat detection by identifying different patterns again there. So in a sense, AI can also help assess security postures of suppliers who are tapping into your ecosystem. So net net, I think AI has a big space to play both on the SOC as well as in the broader security area.

Chad Watt:

Now, on the flip side, we must assume that bad actors are looking for ways to put AI to work in their hacking attacks. Maria, how do you recommend that enterprises and their CISOs prepare for that?

Maria Haight:

So the first thing is data. We still have to go back to the gold of what we have in every company, right? Whether it's sensitive, proprietary, confidential, it's still the asset that needs to be protected. Without good data security, protecting assets via AI just becomes a harder job. But there are ways, right, to limit third-party applications through things like the CASB tools I mentioned earlier, DLP protections, there are things we could be doing. I also think using AI tools within your secured virtual private cloud gives you some freedom without, you know, with limiting the security risk of sensitive data exposure. It's also great to see new companies coming forth that help you securely build and deploy your own agentic AI bots. So those are becoming more of an opportunity for us. But the biggest takeaway for my fellow CISOs is to keep learning. This space is changing so rapidly that it's easy to fall behind or at least feel like you're falling behind and to find a good partner or partners to work alongside. This is not a space that companies should assume they can do well on their own.

Chad Watt:

Rakesh, would you like to add to that?

Rakesh Babu Gollapalli:

Yeah, I couldn't agree more with Maria on that. Bad actors are increasingly sophisticated. They are using AI to start building a lot of different hacking tools. And all these are tools that are available on the dark web, right? In fact, according to a report, over 80% of the phishing emails that you're getting now use AI to mimic humans. And they're being good and better at it on an ongoing basis. In fact, they're using AI to scan code, firmware, and APIs just to identify vulnerabilities. I also heard about AI-powered malware and polymorphic attacks. Now this is where they creating malware that constantly changes. It morphs to evade the anti-virus detection systems that you are installing. So AI is making it very hard and it's making the attacks cheaper, faster, more scalable and more convincing. And I think that's the challenge that CISOs have in the coming world. While AI has its own advantages, unfortunately with bad actors, it can mean a lot of work for CISOs.

Chad Watt:

Let's close out this discussion on the positive. Maria and then Rakesh, what's the most eye-opening work you're seeing with AI right now?

Maria Haight:

I'm going to go broad with this, and I think it's time savings and efficiencies. When we can put very valued people at higher level tasks and get rid of the rudimentary daily grind of tasks that are wasting their time, this is for me where AI just comes to the forefront of being a solution for all of us to find ways to do that. I think every person in their job could find ways that AI can solve their time constraints and put their time to better use. I'll just, you know, from an old IT perspective, test case writing. I mean, just the hours spent on that. And I think code scanning to help our developers feel more confident of what they're putting out, debugging code. These are some great, great opportunities that I think are getting solutions to market faster.

Rakesh Babu Gollapalli:

I would like to share an example from one of my customers, you know, they are fundamentally reimagining their field service business. You know, they're trying to make it become an AI first field service business. So they are bringing the data from their IoT systems, the technical manuals, regulatory information, and making it available via GenAI to transform the way they maintain and manage their install base or their install fleet. Now, this is actually allowing them to reduce the downtime by over 30%, which means more money and more profitability for the customers. But on a philosophical level, the way I think about what AI is doing for us is I think it is allowing us all to go back to being human beings. Now, what do I mean by that? I think, as Maria said, AI is going to help improve productivity. So maybe what it does is it allows us all to work three days a week instead of the five, six days we work and get the same pay. But then with the downtime, what you're doing is you are starting to spend more time with family, probably travel the world more and live a healthier life. So all in all, I think that's a big positive for AI that I hope will come true.

Chad Watt:

Terrific. Thanks so much, Maria, Rakesh. Thank you for joining me for this discussion.

Maria Haight:

Thank you.

Rakesh Babu Gollapalli:

Thank you, Chad. Thank you, Maria.

Chad Watt:

This podcast is part of our collaboration with MIT TechReview in partnership with Infosys. Visit our content hub at technologyreview.com to learn more. Be sure to follow the Infosys Knowledge Institute podcast wherever you get your podcasts. You can find more details in our show notes and transcripts at infosys.com/IKI in our podcast section. Thanks to our producers, Christine Calhoun and Yulia De Bari. Dode Bigley is our audio technician. And I'm Chad Watt with the Knowledge Institute, signing off. Until next time, keep learning and keep sharing.