“While examining the mobile app code, I found a system prompt for an unreleased Gemini integration,” Wong wrote on her blog.
The discovered document, titled Waymo Ride Assistant Meta-Prompt, is a specification of more than 1,200 lines that defines the behavior and boundaries of the in-car AI assistant.
Although the feature is not enabled in public versions of the app, Wong notes that it clearly signals something more advanced than a simple chatbot.
Gemini as an in-car AI assistant
The assistant is designed to answer passenger questions, control limited in-cabin functions such as climate settings, and even calm riders during a trip.
“We can’t share any details. Our team is constantly working to improve features so that Waymo rides are enjoyable, seamless, and helpful. Some of these ideas may be implemented in our system, others may not,” a Waymo spokesperson, Yuliya Ilina, said in a comment.
Alphabet has already integrated Gemini across its technology stack. Waymo previously used large language models’ “world knowledge” to train autonomous vehicles to navigate complex and high-risk driving scenarios.
According to Wong, the assistant is meant to have a clear identity and purpose: a friendly and helpful AI companion whose role is to improve the passenger experience by offering useful information in a safe, calming, and non-intrusive way.
At this stage, Gemini does not have access to certain vehicle controls, including music volume, seat adjustments, route changes, or window controls.
The prompts also define how the bot should respond to questions about competitors such as Tesla, when to end a conversation, and how to avoid speculation about the vehicle’s actions or events during a ride. The assistant can answer general knowledge questions—such as the weather, the height of the Eiffel Tower, or local store hours—but is prohibited from actions like ordering food or making reservations.
Traffic light issues highlight real-world limits
Waymo remains one of the leading players in the autonomous robotaxi market, which it has been developing since 2020. Still, the company continues to face real-world challenges.
On the evening of December 20, 2025, Waymo temporarily suspended robotaxi operations in San Francisco after a widespread power outage left traffic lights offline and autonomous vehicles stalled across the city.
Users shared photos and videos showing clusters of Waymo cars with hazard lights on, blocking traffic and disrupting normal flow.
“The power outage was a large-scale event that caused transportation disruptions throughout San Francisco. Traffic signals were down, and public transit was affected. We are working to adapt our technology to operate during such events,” a Waymo representative said.
Waymo stated that its self-driving system is capable of navigating intersections without functioning traffic lights. However, the scale of the outage significantly slowed decision-making, forcing many vehicles into extended standstills.
Service resumed on December 21, but was paused again three days later for a system update aimed at improving response to future power outages.
“We have always focused on building the Waymo Driver for real-world conditions, including infrastructure failures,” the company said in a blog post.
