Software-defined vehicles (SDVs) are vehicles which can be improved after purchase by using software updates to, for example, add new functions. SDVs use has recently been growing more widespread. While connecting vehicles to the Internet and external services like this offers greater convenience, it also adds new areas that need to be protected from cyberattacks. If a vehicle were to suffer a cyberattack, not only could this prevent the vehicle from being driven safely, but it could also be used to leak or exploit private information, like location information and usage history data. The evolution of AI has also made it easier to deploy attacks. The systems responsible for vehicle cybersecurity need to be able to rapidly identify vulnerabilities and attack indicators, determine their impacts, and implement countermeasures. We talked with three members of Panasonic Automotive Systems, working in the field to protect vehicle safety, about how they use AI and what roles human personnel are responsible for.
Vehicles are now also targets.
The changing face of cyberthreats in the AI era
To start off with, could you talk about why vehicles need cybersecurity and how advances in AI have changed the threats that are out there?
Ishii
Modern vehicles wirelessly connect to base stations, just like smartphones do, and send and receive data via the Internet. In other words, in a sense, in some ways they are like “smartphones with driving functions.” Because of that, it is of course possible that they will be targeted by cyberattacks. Over a decade ago, researchers had demonstrated in the lab that it was possible to interfere with car behavior remotely.
Ito
Potential risks can be broadly divided into the four categories of safety, financial, operational, and privacy impacts. If vehicle operations can be affected, that directly affects the safety of the people onboard. If an attacker stole location information, usage history, transaction information or the like, the damages could extend beyond the vehicle itself. Attackers could even electronically unlock and steal the entire vehicle. That’s why cybersecurity is essential for vehicles, too.
Ishii
As AI has advanced, it’s gone beyond just chat. Modern AI can search for information on its own, write its own code, and use various tools. These functions can also be used in cyberattacks, so the barrier to launching a cyberattack has been lowered. In the past, an attacker had to have some level of knowledge or skill. Now, AI can aid in gathering information and creating programs, which has made it easier for even people without any specialized knowledge to try launching attacks.
Ito
The use of AI in attacks isn’t the only problem. The systems within vehicles also have AI embedded in them, so the AIs themselves can become new attack targets. As an example, one attack vector that has been researched is making changes that are imperceptible to the human eye in vehicle camera images to cause AI systems to make mistakes in recognizing street signs. Not only will the number of attacks using AI increase, but the number of AI that need to be protected will also increase. On the defense side, we need to address these problems even more quickly and create defenses that anticipate attacks.
So when you protect the security of vehicles, what kind of “secure state” are you protecting?
Ito
To put it succinctly, it’s the state in which the vehicle’s basic operations, like driving, turning, and stopping, all function correctly and services supplied by auto manufacturers and software manufacturers work the way the user intends them to. I think a secure vehicle is one in which external intrusion is prevented and the vehicle’s functions and services can be used with peace of mind.
Protecting vehicles from three perspectives,
from their initial configuration
until after they are shipped
How do each of you protect security in each of the stages of the vehicle lifecycle? Could you explain and also describe your role in this work?
Tomiie
I develop tools for identifying what threats there are during the vehicle and system concept phase.
During this stage, we don’t even have program source code, let alone a physical vehicle. We identify
the vehicle’s structure, its communications, and the data it uses. Then we comprehensively analyze
potential attack paths and threats such as information leaks or tampering.
We evaluate each
of these risks and present the developers with the measures that need to be taken in the form of
cybersecurity requirements. This process is called “TARA”*1. We discover risks before
actual production begins and we build countermeasures into the vehicle’s very design. Addressing
problems early-on, instead of later in the process, significantly reduces the risk of having to make
major design changes.
That said, design changes happen multiple times during the
conceptualization stage, too. Performing analysis again each time would place a significant burden
on us, and would result in deviations in analysis results depending on the level of experience of
the people performing the analysis. That’s why we’ve automated TARA and developed a
solution*2 that can be used by anyone, not just security experts. This reduces the amount
of time this work takes and produces more consistent analysis results.
- *1 TARA: Threat Analysis and Risk Assessment. In this method, potential system threats are identified, risks are analyzed, and cybersecurity requirements are derived and implemented in designs.
- *2 Solution: The “VERZEUSE® for TARA” ISO/SAE 21434-compliant risk analysis solution. A test version of this solution has been available since May 2026, before the launch of the full official service. (VERZEUSE® is a registered trademark of Panasonic Automotive Systems Co., Ltd. or its affiliates in Japan. It is a neologism that combines the Spanish word “ver,” which means “see,” and “Zeus,” the god. It represents looking down from the heavens like a god to watch over the safety of society.)
Ishii
I check if any vulnerabilities remain before products are shipped, so I perform verification
by performing security testing, including penetration testing, from an attacker’s perspective. Using
products that are almost in their completed state, I use external connections like Wi-Fi or
Bluetooth to see if unauthorized access is possible.
I don’t just look at whether the attacks
succeed or fail. I also determine why we arrived at the results we did, what kinds of problems there
were in the background, what kinds of impacts there would be on products and users, and the
priorities of the risks to be addressed. I separate the feasibility of an attack from the impact
that an attack has when successful, and consider each separately. Some attacks may be relatively
easy to mount but have limited impacts. Others may be more difficult, but could threaten people’s
lives. Based on these, I determine the priority we should use in addressing them.
Ito
I’m engaged in exploratory research focused on how to combine AI with security over the medium and
long term. I consider how and where AI should be used, from development processes to shipping and
beyond.
We keep on implementing security measures even after products have shipped. New
vulnerabilities are being found every day, and the methods used in attacks are evolving. We monitor
and analyze information and logs from vehicles after they’ve shipped. If we discover problems, we
restore the vehicles to a safe state by, for example, updating their software. Vehicles need to keep
being protected throughout their long lifespans, all the way until they are ultimately scrapped.
The work that can be left up to AI
and the judgment that humans are responsible
for
As people responsible for protecting the safety of vehicles, how do you use AI? What areas of security are people responsible for?
Ito
AI excels at working with massive amounts of information. It wouldn’t be feasible for people to
deal with all the vulnerability-related information that we work with every day and the massive logs
we collect from vehicles. We use AI to read and analyze that data and summarize it in a report,
which reduces the amount of time it takes for people to make judgments. If attackers are going to
use AI to speed up their attacks, then us defenders also need to use AI to speed up our response
cycle.
There are also aspects of threat analysis that can be left up to AI. One of these is
on the input side, using AI to read large volumes of specifications and organize the information
needed to perform analysis. Another is on the output side, rephrasing analysis results and security
requirements in ways that can be easily understood even by developers who don’t have detailed
security knowledge and organizing what steps should be taken next. It can serve almost like an
interpreter, connecting technical analysis with the development field.
But we can’t leave
everything, even the design of the core software that supports analysis, entirely up to AI.
Sometimes AI produces different output from the same input. When you’re performing analysis and
verification that people’s lives and safety depend on, it is vital that you can reproduce the same
results under the same conditions. That’s why we have people who understand objectives and risks
design rules and logic and carefully carry out verification work. That ensures consistency and
reliability. The flexible inferencing and decision-making support of AI, together with
reproducibility of software designed and verified by people, combine to create an analysis platform
that is both efficient and reliable.
Tomiie
With AI, even people with little experience can produce countermeasure proposals that are close to those produced by veterans. However, people need to confirm if those proposals are actually correct and if they’re really needed for the systems in which they are to be deployed.
Ishii
I don’t currently use AI in evaluation work myself, but I think that AI could improve efficiency for repetitive, routine testing. But that leaves the question of how to interpret the results. People are essential to going beyond determining whether or not an attack was successful to explore the background behind it and what actual impact it would have. We need a realistic perspective to determine if an attacker would really apply such a difficult attack process. Even if we can leave experimentation and security testing up to AI, ultimate responsibility lies with people and organizations.
Ito
In using AI, what’s important is clearly distinguishing between what can be left up to AI and what decisions people should be responsible for, based on the impact of those decisions. While decisions with minor impacts could be left up to AI, decisions with major impacts need to be made by people after confirming their contents and the evidence. Even if we use AI, ultimately the responsibility lies in our people and our organization.
Constantly updating protection
to deliver constant and consistent security
What strengths does your company have when it comes to ensuring vehicle security in the AI era?
Ishii
The people in our company have experience in a wide range of fields, like semiconductors, electronic components, and mobile phones. We have people who understand not only how software works at the structural level, but also hardware—people who have been involved in the development of different kinds of products. That’s why they can apply a wide range of perspectives to security.
Tomiie
One time, there was a hardware-side measure which, at first glance, didn’t seem to be particularly
related to what we were working on. One of our veteran personnel, who had previous experience with
semiconductors, explained just why the measure was needed. If it had been just me, with my software
knowledge alone, I might have decided it was unnecessary, but thanks to his knowledge regarding
semiconductors, we were able to connect the dots.
I think one of our strengths is that we have
both expertise as a software vendor and expertise as a supplier that directly supplies components to
automobile manufacturers. We understand the relationships between hardware and software, and we have
environments that enable us to perform confirmation by trying attacks on actual devices. And, above
all else, we fully understand that people will be riding in those vehicles. I think that experience
has provided us with strengths when it comes to protecting vehicle security.
What do you continue to place great value on in the face of the changes being brought about to security by AI?
Ito
AI is just a tool. We were protecting people before AI, and we’ll keep protecting people now that AI is here. Even if attack methods become more diverse and the scope requiring protection grows, users still need to be able to enjoy peace of mind as they use their vehicles’ basic operations of driving, turning, and stopping, and the services offered by their vehicles. The way we protect them may change, but the objectives of our work do not.
Ishii
Dealing with the growing number of attacks places a significant burden not only on departments involved in security but also those in the field who develop new products. That’s why we’re using AI to improve the efficiency of work that can be left to AI and having people make decisions about which risks need to be prioritized based on product specifications and how products are used. Going forward, I think it will become even more important to reduce the workload placed on those in the development field while reliably ensuring the required level of security quality.
Ito
Ultimately, it all comes down to the safety of the end user. We’re a Tier 1 supplier, and our direct customers are automobile manufacturers, but we never forget that beyond them lie the people who actually ride in the vehicles, and we always think about how to enable them to use their vehicles while enjoying peace of mind. That’s something that hasn’t changed at all, and all three of us here feel the same about it.
Takayoshi Ito
Joined the company in 2018 and was assigned to the Security Development Department. Ever since, he has been engaged in exploratory research with a medium- and long-term perspective and he has worked to refine security measures that leverage AI, led by his mission to create next-generation security technologies that support the safety and security of mobility. Believing that “security should work in the background,” his goal is to create a society in which everyone can experience the value of mobility while enjoying peace of mind.
What brings him comfort?
Looking at the face of his sleeping dog.
Kotone Tomiie
Joined the company in 2022 and was assigned to the Security Development Department. Ever since, she has been involved in the development of tools that automate and support threat analysis and risk assessment (TARA) during the initial stages of vehicle and system design. In addition to automating analysis work and reducing workloads, she also makes improvements to tools that are used in a wide range of companies. Her goal is to manage risk through the entire product lifecycle, starting with initial design, and create a safe, secure mobility society.
What brings her comfort?
Brewing tea.
Kou Ishii
Joined the company in 2018 and was assigned to the Security Development Department. Ever since, he has been responsible for launching security evaluation services that simulate real-world attacks against products before shipping and verify whether or not they have any vulnerabilities. He is also involved in developing frameworks for verifying the actual implementation of security functions. He always stays a step ahead in identifying vulnerabilities and developing countermeasures with the goal of establishing systems for protecting mobility and ensuring that it is safe and comfortable.
What brings him comfort?
Coming up with an elegant solution to a difficult puzzle.
