Google on the privacy of autonomous AI assistants

At the 2026 Global Summit of the International Association of Privacy Professionals (IAPP), the world's largest annual gathering of digital responsibility professionals, Kent Walker, President of Global Affairs at Google and Alphabet, delivered a keynote speech on the future of data protection. The fundamental context for this discussion was an unprecedented technological leap: modern AI models have become three hundred times more efficient compared to cutting-edge developments from just two years ago. This is not a three hundred percent improvement, but a manifold increase in fundamental capabilities. Today's algorithms have gone far beyond simple prediction.

In this context of rapid progress, consumer expectations are fundamentally changing regarding how exactly technology should integrate into their lives. The company's founders, Larry Page and Sergey Brin, always envisioned search engines evolving from simply providing answers to active suggestions and then to actual assistance, and now, for the first time, the industry has reached the stage of practically realizing this vision. Ordinary users no longer want to interact with basic chatbots; they need a reliable, trusted assistant that connects disparate facts and takes on routine tasks. A prime example of this transition was the recent launch of the personal intelligence feature in the AI-powered search engine in the United States.

Despite innovations enabling the long-held idea of providing everyone with a personal assistant, the developers' approach to implementing these powerful tools remains focused exclusively on people. The strategy for deploying new autonomous capabilities relies on caution and safety. Initially, the company engages a limited group of trusted testers to deeply understand the real needs and audience reactions to the functionality. Only after receiving comprehensive feedback does a responsible, gradual scaling of services to a wider public begin. At each stage of this process, data is collected on whether users like the tool and if it brings them tangible benefits.

Successful examples of implementing such deep personalization at the state level already demonstrate the enormous potential of autonomous systems when their architecture is approached correctly. In particular, Ukraine, which has become one of Europe's most digitally advanced countries, in cooperation with technology specialists, has created the national AI assistant Diia.AI. This service goes far beyond traditional reference systems and provides government services tailored to the individual needs of each citizen, directly within a familiar chat interface.

However, the key factor for the mass adoption of such technologies remains the unconditional trust of users, who always want to stay in control and fully manage the situation. This requires close cooperation between industry representatives and regulatory bodies to create context-dependent protections. Developers must carefully assess in which scenarios friction when interacting with AI is justified, and in which it is not. For example, if a person entrusts their agent with purchasing a specific item, they expect seamless task execution: the use of relevant payment data and addresses without interruptions or pop-ups.

To ensure these guarantees at the development level, three key methods are applied to preserve people's right to self-determination regarding their data. Firstly, direct controls over agent access are implemented, allowing users to easily enable and disable connections between applications. Secondly, strict system limitations are established for sensitive areas; for example, the Gemini model fundamentally avoids proactively making assumptions on potentially sensitive topics. Thirdly, AI agents are trained exclusively on the volume of data that is strictly necessary to provide a quality service and enhance its usefulness over time.