Google’s newly launched AI agent, Gemini Spark, demonstrates advanced automation by creating a detailed birthday party plan using extensive access to a user’s personal emails, calendar, and documents, yet it failed to correctly identify the user’s boyfriend as more than a close friend.
What happened
Introduced at Google’s recent I/O developer conference, Gemini Spark is an AI assistant that accesses users’ personal data to automate tasks like scheduling, emailing, and internet browsing. A beta version is now available to subscribers of Google’s AI Ultra plan, priced from $100 a month.
In a hands-on test, a user granted Gemini Spark full access to their Gmail, Google Docs, and Calendar to plan their birthday party. The AI quickly produced a comprehensive five-page itinerary including a guest list, venue rules, nearby dining options, and after-party plans. The guest list and venue details were sourced directly from the user’s emails and travel history.
Despite analyzing the data thoroughly, Gemini Spark referred to the user’s live-in boyfriend as merely a “close friend and frequent companion,” repeatedly failing to recognize the actual nature of their relationship. Moreover, the AI excluded the user themselves from the guest list. The AI’s attempts to book dinner reservations via a remote browser also encountered technical issues.
The AI’s after-party suggestions featured exclusively LGBTQ+ venues, which it explained was based on explicit references found in the user’s Google Workspace files rather than inferred personal identity.
Google warns users about risks associated with granting an AI this level of access, highlighting potential prompt injection attacks where malicious actors could exploit the AI’s permissions to expose sensitive data or conduct unauthorized actions.
Why it matters
Gemini Spark’s ability to automate complex personal tasks by tapping into detailed user data marks a significant advancement in AI-powered digital assistants. However, the agent’s failure to understand nuanced human relationships underscores current limitations in AI’s contextual and emotional comprehension.
The security warnings issued by Google emphasize the ongoing concerns about privacy and data vulnerability when allowing AI systems unrestricted access to personal information. This balance between convenience and security will be crucial as AI agents become more integrated into daily life.
Background
Gemini Spark represents Google’s response to early 2026 innovations like the viral OpenClaw AI assistant, which similarly personalized task automation but was criticized for occasional mishaps related to privacy and understanding user intent. Google’s Gemini chatbot framework powers Spark, available on both desktop and mobile platforms, including iPhones.
Google classifies user commands to Spark as “tasks” rather than prompts, highlighting its goal of a more interactive and ongoing assistant designed to operate independently with user approval on sensitive actions like sending emails.
Sources
This article is based on reporting and publicly available information from the following source:
Read more Artificial Intelligence stories on Goka World News.