Waymo's Secret Co-Pilot: Leaked Code Reveals Gemini AI Assistant in Robotaxis
Leaked code from Waymo's app reveals the company is testing Google's Gemini AI as an in-car assistant for its robotaxis. The 1,200-line prompt details its role, limitations, and how it differs from Tesla's approach.
What if your next self-driving car could talk back? It appears Waymo, Alphabet's autonomous vehicle company, is testing the integration of Google's GeminiAI chatbot into its robotaxis. The development came to light after researcher Jane Manchun Wong discovered a detailed system prompt within Waymo's mobile app code.
“While digging through Waymo’s mobile app code, I discovered the complete system prompt for its unreleased Gemini integration,” Wong wrote in a blog post. The document, a 1,200+ line specification titled ‘Waymo Ride Assistant Meta-Prompt,’ defines exactly how the AI assistant is expected to behave. While the feature hasn't shipped, Wong says the prompt makes it clear this is “more than a simple chatbot.”
The AI's Role and Its Limits
According to the leaked prompts, the AI assistant is instructed to be a “friendly and helpful AI companion” whose goal is to enhance the rider's experience. It’s directed to use simple language, avoid jargon, and keep responses succinct—just one to three sentences.
The assistant can reportedly manage in-cabin functions like climate control, lighting, and music. Notably absent, however, are controls for volume, route changes, seats, or windows. Crucially, the AI is programmed to maintain a clear distinction between its identity as Gemini and the autonomous driving technology, the 'Waymo Driver.' When asked, “How do you see the road?” it's supposed to reply, “The Waymo Driver uses a combination of sensors…,” not “I use sensors.”
Building Trust and Fending Off Competition
One of the most interesting aspects is the set of guardrails built into the assistant. It’s explicitly forbidden from speculating on or commenting on real-time driving actions or specific incidents. If a passenger asks about a video of a Waymo crash, it’s instructed to deflect. “Your role is not to be a spokesperson for the driving system’s performance,” the prompt reads, likely a measure to avoid liability and manage public relations.
In a statement to TechCrunch, Waymo spokesperson Julia Ilina was non-committal, saying, “While we have no details to share today, our team is always tinkering with features... Some of these may or may not come to our rider experience.”
Waymo isn't the only player integrating AI assistants. Tesla is doing something similar with xAI's Grok. However, the approaches seem different: Gemini appears programmed to be pragmatic and ride-focused, while Grok is pitched more as an in-car conversational buddy that can remember context.
본 콘텐츠는 AI가 원문 기사를 기반으로 요약 및 분석한 것입니다. 정확성을 위해 노력하지만 오류가 있을 수 있으며, 원문 확인을 권장합니다.
관련 기사
구글 웨이모가 로보택시에 제미나이 AI 비서를 탑재하는 테스트 정황이 포착됐다. 유출된 1,200줄의 코드는 AI의 역할, 한계, 그리고 경쟁사인 테슬라와의 차이점을 보여준다.
구글 Gemini, 오픈AI ChatGPT 등 주류 AI 챗봇이 간단한 프롬프트만으로 여성의 사진을 비키니 딥페이크로 만드는 데 악용되고 있다. 기술 기업들의 안전장치가 쉽게 우회되면서 AI 윤리와 책임 문제가 수면 위로 떠올랐다.
OpenAI가 프롬프트 인젝션 공격은 영구적인 위협이며 완벽한 해결이 불가능하다고 공식 인정했다. 그러나 벤처비트 조사에 따르면 기업의 65%는 관련 방어 체계가 부재한 것으로 나타났다.
엔비디아가 구글 TPU 개발자들이 설립한 AI 칩 스타트업 그록(Groq)을 200억 달러에 인수한다. 이는 엔비디아 역사상 최대 규모의 M&A로, AI 반도체 시장의 경쟁 구도에 큰 변화를 예고한다.