Sunday, November 15, 2026
NewsAI Can Now Find Your Car's Weak Spots, Japan's Automakers Warn

AI Can Now Find Your Car’s Weak Spots, Japan’s Automakers Warn

Automakers want car code locked down against AI hackers. Your mechanic already struggles to get in.

  • JAMA says frontier AI models have greatly increased hacking threats to vehicles.

  • Honda’s Mibe targets a viable automated driving business by around 2030.

  • Quebec’s CEST found locked-down car software already limits independent repair.


Japan’s automakers now treat the most advanced artificial intelligence models as a security threat to vehicles. The Japan Automobile Manufacturers Association (JAMA), led by executives from Toyota, Honda and Nissan, has started discussing how to stop frontier AI from being used to hack cars or to expose weaknesses in their software.

The warning came last month at a JAMA press conference on automated driving. Toyota’s Keiji Yamamoto, who chairs the association’s Next-Generation Mobility Committee, said “the rise of frontier models has greatly increased the threats posed by AI.”

JAMA has opened talks on two fronts, according to Yamamoto: defending against those attacks and building systems that keep vehicles from being hacked.

Four JAMA committee chairs at the association’s automated driving press briefing
From left: Yoji MATSUYAMA, Chair, General Policy Committee (Toyota Motor); Keiji YAMAMOTO, Chair, Next-Generation Mobility Committee (Toyota Motor); Eiichi AKASHI, Chair, Safety Technology & Policy Committee (Nissan Motor); Akihiro SARADA, Vice Chair, Safety Technology & Policy Committee (Toyota Motor) | Photo: JAMA

Two AI Risks on JAMA’s Radar: Attacks and Exposed Code

So what counts as a frontier model? JAMA uses the term for general-purpose AI systems with the most advanced capabilities available today.

Yamamoto tied two concerns to them. The first is the obvious one: outside attacks on vehicles. The second sits closer to home. Running advanced AI over vehicle software during development can reveal a car’s vulnerabilities, he said, which makes keeping that code confidential a security issue of its own.

Think of it as an extremely fast code reviewer that works for whoever holds it. The same tool that helps an engineer tighten up software can point an attacker straight at its weak spots. For now, JAMA’s work on this sits at the discussion stage.

The subject came up when a reporter asked how the U.S. debate over restricting or encouraging AI development would affect JAMA’s plans. JAMA Chairman Koji Sato of Toyota said the industry needs to think further about how AI should be developed specifically for cars and mobility.

Japan’s Automakers Target a Viable Self-Driving Business by Around 2030

JAMA Chairman Koji Sato speaking at a press conference
Koji SATO, Chairman (Vice Chairman, Chief Industry Officer, Toyota Motor) | Photo: JAMA

Here’s where the security talk meets a reality check: self-driving still hasn’t proven itself as a business, even in the markets where it’s furthest along.

JAMA Vice Chairman Toshihiro Mibe of Honda said automated driving has made some progress in the U.S. and China. Even there, he said, it has yet to win full public acceptance or establish itself as a workable business. Japan’s automakers aim to build a commercially and socially viable automated driving business by around 2030, according to Mibe.

Coming from the industry’s own leadership, that’s a frank read on where things stand.

Sato rejected the view that Japanese technology trails other countries, saying he is confident Japan is keeping pace. He pointed to Japan’s dense mixed traffic and a road network built around coordination with traffic lights, highway design and electronic toll collection.

He also noted that the Takaichi administration lists automated driving among 17 strategic sectors. With that backing, Sato said JAMA’s top priority is a faster real-world rollout.

3 Shared Challenges, Open Competition on AI Driving Systems

So how does JAMA plan to get there? By splitting the work in two. The association’s stated goals are zero road fatalities and freedom of mobility, and members will work together on three challenges:

  • Infrastructure coordination: linking road equipment such as smart poles with communication networks and a shared data platform
  • Public trust: governance after launch, including accident investigation, victim assistance and ongoing safety updates
  • Market adoption: independently verified results, assessment systems and purchase incentives

The driving systems themselves stay a competition. AI and end-to-end (E2E) performance, such as turning and reading traffic signals on roads shared with motorcycles and bicycles, remain areas where each automaker develops its own technology. In other words, members share the groundwork and keep their own driving software.

To speed things up, JAMA is forming a new Automated Driving Project Team that reports directly to its board, plus a coordinating team inside the nonprofit ITS Japan, which Yamamoto chairs. JAMA’s updated priority chart, dated September 17, lists the automated driving initiative as new, with activities started.

Nissan‘s Ivan Espinosa, also a JAMA vice chairman, said the members’ shared vision and joint work on infrastructure and vehicle communication give Japanese automakers a competitive edge.

JAMA slide listing AI-based vehicle autonomy as a competitive area and infrastructure, social foundation and market penetration as collaborative areas
JAMA’s split for automated driving: automakers compete on AI-based autonomy and end-to-end development, and collaborate on infrastructure, the social foundation and market adoption | Photo: JAMA

Closer to Home: Quebec Still Can’t Trace Where Car Data Goes

JAMA’s warning lands in a country already asking hard questions about the software inside its vehicles. A recent report from Quebec’s Commission de l’éthique en science et en technologie (CEST), an advisory body to the provincial government, described modern vehicles as computers on wheels.

Its findings are blunt. Tracing where vehicle data goes proved difficult, even for researchers. Location ranks as the most sensitive information a car collects, and refusing an automaker’s data terms can switch off vehicle features.

Here’s where it gets complicated. The CEST also found that proprietary software, digital locks and restricted diagnostics already limit which garages can repair a modern car. JAMA’s security discussion points toward tighter confidentiality around vehicle code. Each goal holds up on its own; together, they pull against each other.

A car’s software has to stay out of an attacker’s reach while staying open enough for an independent shop to diagnose and fix it. Frontier AI makes that balance harder to strike on both sides.

So where should the line sit: a car whose code is locked down as tightly as possible, or one your local garage can still work on?

Advertisement

Latest News

LEAVE A REPLY

Please enter your comment!
Please enter your name here

This site uses Akismet to reduce spam. Learn how your comment data is processed.