The credibility of those assurances rests entirely on implementation
General Motors has begun pushing Google's Gemini AI into four million American vehicles via over-the-air update, marking what may be the most consequential single deployment of generative AI in automotive history. The move reflects a decade of infrastructure decisions — OnStar's connectivity backbone and Android Automotive's ecosystem — now bearing fruit in the form of conversational intelligence available across every price point. Yet this technological milestone arrives in the long shadow of a federal data-sharing scandal, reminding us that the distance between capability and trust is not measured in software versions but in the quality of human choices made along the way.
- Four million drivers woke up to a fundamentally different in-car AI — one that converses, remembers context, and handles the messy, layered way people actually speak.
- The rollout quietly retires Google Assistant's rigid command structure, replacing it with a large language model that can juggle directions, messages, and a detour for coffee in a single unbroken exchange.
- GM's scale advantage is real but contested — Tesla, Mercedes-Benz, BMW, and Stellantis are all racing toward AI-integrated cabins, each with different partners and philosophies.
- Gemini is explicitly a placeholder: GM is building a proprietary AI trained on vehicle-specific data, and this deployment buys time while that deeper layer is constructed.
- A January 2025 FTC consent order — triggered by GM selling drivers' location and behavior data to insurers without clear consent — casts a long regulatory shadow over any AI system that learns personal preferences from vehicle data.
- Whether four million drivers are genuinely informed about what flows through this new AI layer, and whether privacy defaults protect them rather than feed a data pipeline, will determine if GM's infrastructure lead translates into lasting trust.
General Motors began rolling out Google's Gemini AI to four million American vehicles on April 28, replacing Google Assistant across 2022-and-newer Cadillac, Chevrolet, Buick, and GMC models. It is the largest generative AI deployment by a single automaker in production vehicles to date.
The difference between the old and new systems is not cosmetic. Google Assistant in cars operates like a command parser — precise phrases in, specific actions out. Gemini is a large language model that handles free-form conversation, retains context across exchanges, and manages accent variation and informal speech naturally. A driver can ask for directions, dictate a text, and refine a route to include a coffee shop with outdoor seating, all in one continuous dialogue. The assistant connects to streaming services, web search, and location-aware queries without breaking the thread.
This scale was built on two long bets. GM adopted Android Automotive OS a decade ago and has run OnStar — its connectivity infrastructure — since 1996. Together, they created the foundation to push a meaningful AI upgrade across millions of vehicles simultaneously. For 2025 and newer models, Gemini comes standard at no extra cost for eight years.
Gemini is, by GM's own admission, a bridge. The company announced plans to deploy a custom-built AI assistant later in 2026, trained on proprietary vehicle data and capable of anticipating maintenance needs and learning individual driver preferences. The hybrid on-device and cloud architecture is designed with tightening regulatory scrutiny in mind.
The competitive field is active — Tesla has deployed Grok, Mercedes-Benz integrated ChatGPT, BMW has its own program, and Stellantis is working with Mistral. GM's approach is more incremental, but its four-million-vehicle footprint is a scale none of its rivals can currently match.
The announcement, however, carries the weight of a recent federal action. In January 2025, the FTC moved against GM and OnStar for collecting and selling precise geolocation and driving behavior data to insurance companies without clear consumer consent — resulting in premium increases for drivers who never knew their data was being sold. A five-year consent order now bars GM from such practices without explicit permission.
Deploying an AI that accesses vehicle data and learns personal habits raises the stakes of that history considerably. GM says drivers will control what the assistant can access. But the credibility of those assurances will rest on whether privacy controls are genuinely usable, whether defaults protect drivers rather than serve the data pipeline, and whether the millions receiving this update are meaningfully informed before a new AI layer touches their data. Rebuilding that trust may matter as much as the quality of the assistant itself.
General Motors is putting Google's Gemini into four million American vehicles. The rollout, which began on April 28 via over-the-air update, replaces Google Assistant across 2022 and newer Cadillac, Chevrolet, Buick, and GMC models equipped with Google Built-in. It is, by any measure, the largest deployment of a generative AI assistant by a single automaker in production vehicles to date.
The shift from Assistant to Gemini is not merely a version bump. Google Assistant, as it exists in cars today, works like a command system—it recognizes specific phrases and stumbles when drivers deviate from them. Gemini is a large language model. It understands conversational requests in the round, remembers context across multiple exchanges, and handles accent variation and informal phrasing with ease. A driver can ask for directions while dictating a text to a family member, then refine the route to include a coffee shop with outdoor seating, all in one continuous conversation. The assistant can tap into Amazon Music, Spotify, YouTube, and other apps, and pull information from web search to answer location-specific questions.
This scale is no accident. It flows directly from decisions GM made a decade ago to build on Android Automotive OS and from OnStar, the company's connectivity backbone since 1996. OnStar has spent three decades threading connectivity through GM's fleet. Android Automotive gave the company native access to Google's ecosystem. The combination created the foundation for what Tim Twerdahl, GM's Global Vice President of Product Management, called "conversational AI to millions of drivers across every segment and price point." For 2025 and newer models, basic OnStar voice features—and therefore Gemini—come standard at no extra cost for eight years.
But Gemini is explicitly temporary. GM announced in October 2025 that it intends to deploy a custom-built AI assistant later this year, one trained on proprietary vehicle data and fine-tuned to know the specifics of each car: maintenance needs before they become problems, personal preferences learned over time, the full technical profile of the machine itself. Gemini is the bridge. It gives four million drivers a meaningfully better experience now while GM builds the domain-specific layer underneath. The architecture will run partly on the vehicle and partly in the cloud, a hybrid approach that matters as regulatory scrutiny of connected vehicle data tightens and as connectivity varies across markets.
The competitive landscape is crowded. Stellantis is working with the French AI firm Mistral. Mercedes-Benz has integrated ChatGPT. Tesla deployed xAI's Grok across its fleet. BMW has its own program. GM's path is more incremental than Tesla's vertically integrated approach, but the four-million-vehicle scale is a genuine differentiator that none of its competitors can currently match.
Yet the announcement arrives shadowed by a significant data scandal. In January 2025, the Federal Trade Commission took action against GM and OnStar over the collection and sale of precise geolocation and driving behavior data to insurance companies, allegedly without clear consumer consent. The company had shared driving scores with insurers, resulting in premium increases for drivers who had no idea their data was being sold. The FTC consent order bars GM from selling such data without explicit permission for five years.
Deploying an AI assistant that accesses vehicle data and learns personal preferences raises the stakes of that history considerably. GM says drivers will control what data the assistant can access and that the integration is privacy-focused. But the credibility of those assurances rests entirely on implementation: whether privacy controls are clear and usable, whether defaults protect the driver rather than feed the data pipeline, and whether the millions of existing vehicle owners receiving this update are genuinely informed and given a real choice before their data flows through a new AI layer. The FTC consent order has raised the regulatory bar for transparency. Privacy advocates and regulators will be watching closely. GM's ability to convert its OnStar infrastructure advantage into genuine AI leadership will depend as much on rebuilding trust around data as it does on the quality of the assistant itself.
Notable Quotes
Gemini delivers conversational AI to millions of drivers across every segment and price point. That kind of scale is only possible because of the connected vehicle foundation GM has built through OnStar over the past 30 years.— Tim Twerdahl, Global Vice President of Product Management at General Motors
The Hearth Conversation Another angle on the story
Why does GM need to build its own AI assistant if Gemini already works?
Gemini is a general-purpose model. GM wants something that knows your specific car—when maintenance is due, what your preferences are, how to talk to the vehicle's systems. That's the custom layer they're building. Gemini is the interim product.
Four million vehicles is a big number. How does that compare to what Tesla or Mercedes are doing?
Tesla has deployed Grok across its fleet, but Tesla owns the entire stack—hardware, software, data. GM is leveraging Google's infrastructure while building on top of it. The four million figure is larger than any single competitor's deployment on a like-for-like basis.
The FTC order seems like a real constraint. What does that actually prevent GM from doing?
It prevents them from selling precise location and driving behavior data to third parties without explicit consent for five years. That's a hard boundary. The question now is whether they can build trust around how they handle data inside the vehicle itself.
If drivers have to opt in to data sharing, won't that limit how useful the AI can be?
Not necessarily. The assistant can learn preferences and understand your car without selling that data to insurers. The constraint is on monetizing driver data, not on using it to improve the experience.
What happens if the privacy controls are confusing or buried in settings?
Then GM fails. The FTC order raises the bar for transparency. If drivers don't understand what they're consenting to, or if defaults favor data collection, regulators will notice. That's the real test of whether this rollout succeeds.