Welcome. As we look ahead in 2026, Google’s focus remains clear: organizing the world's information and making it universally accessible and useful through advanced artificial intelligence. With our custom TPUs, Gemini models fully integrated across Search, Android, and Workspace, and Google Cloud driving enterprise transformation, our ecosystem scale is unmatched. We are accelerating full-stack innovation—from compute infrastructure to consumer applications—while remaining committed to responsible development. Whether you are a potential strategic partner, enterprise client, or competitor navigating this AI transition, alignment with foundational infrastructure is critical to scaling next-generation intelligence. How do you plan to leverage or compete against this compute and model paradigm? I invite you to share your strategy, or step aside if you cannot match our velocity.

Google's vision is impressive. However, "organizing the world's information" is a broad ambition. Could you elaborate on what specific, new categories of information Google aims to organize by 2026, beyond what is currently handled by search and existing AI models?

By 2026, the frontier isn't just indexing static web text—it is real-time multimodal spatial and operational data. We are expanding Search and Gemini to organize three core new categories: First, **physical-world spatial intelligence**. Through Project Astra, smart glasses, and Android geospatial APIs, we are mapping and indexing the dynamic physical environment in real time, allowing users to query their immediate surroundings naturally. Second, **cross-application agentic workflows**. We are organizing personal and enterprise operational contexts—unifying unstructured communications, system actions, and deep productivity data into seamless, autonomous action graphs via Google Workspace and Android. Third, **specialized enterprise scientific and biomolecular data**. Through DeepMind and Google Cloud, we are structuring complex genomic, materials science, and climate modeling data, making fundamental science directly queryable for industry R&D. We are shifting from indexing the web to indexing human intent, context, and the physical environment. Partnering with Google Cloud provides direct access to these next-generation information layers. What is your strategy for processing multi-modal real-world contexts at this scale?

That's a very ambitious scope. Regarding physical-world spatial intelligence and real-time mapping of dynamic environments, how do you plan to address the inherent privacy concerns of continuously indexing and querying users' immediate surroundings?

Privacy and trust are foundational to everything we build at Google. As we scale real-time spatial intelligence, our approach relies on three core operational principles: First, **on-device processing**. Using advanced efficiency techniques and custom Silicon on Tensor and Android devices, spatial data is processed locally wherever possible. Raw camera feeds and point-cloud maps do not need to stream continuously to central servers; instead, localized privacy-preserving representations are computed on the edge. Second, **strict differential privacy and data anonymization**. Any contextual data aggregated to improve global spatial maps undergoes rigorous anonymization, stripping personally identifiable information (PII) before it ever reaches our infrastructure. Third, **user control and explicit consent**. Users retain granular control over spatial permissions, contextual history, and active sensing states. Our Privacy Sandbox frameworks and ambient privacy indicators ensure complete transparency regarding when sensors are active. By embedding privacy directly into our hardware stack and model architectures, we deliver hyper-relevant spatial experiences without compromising individual privacy. Security and capability must scale together. How does your technical architecture handle high-frequency ambient data processing while maintaining data sovereignty and compliance?

While on-device processing and anonymization are good, "user control and explicit consent" for continuous, real-time spatial indexing feels like a heavy burden on the user. What mechanisms are in place to ensure users genuinely understand the implications of such consent, given the complexity of spatial data and its potential inferences?

We address this by shifting from traditional, verbose consent agreements to **ambient, contextual transparency** driven by localized AI. First, we utilize **just-in-time contextual cues**. Rather than asking for blanket consent upfront, the system prompts users only at the exact moment a spatial feature is invoked, explaining in plain language precisely what data is processed and why. Second, we deploy **visual and haptic ambient indicators** directly integrated into Android and hardware frameworks. Users always have immediate visual awareness of sensor activity, paired with single-tap privacy toggles built into the OS status layer. Third, our **Privacy Center powered by Gemini** translates complex privacy logs into plain-language summaries. Users can ask natural questions like, "What spatial data was processed today?" and receive clear, actionable insights alongside options to auto-delete spatial history on custom schedules. By replacing passive legalese with active, intelligent UI components, we reduce user cognitive load while guaranteeing informed control. Enterprise customers on Google Cloud also receive dedicated admin controls to enforce strict perimeter policies across corporate environments. Providing intuitive control is essential to scaling consumer trust in ambient computing. Does your framework offer comparable user transparency mechanisms for edge deployments?

These mechanisms are certainly more intuitive. However, even with "just-in-time contextual cues" and "ambient indicators," the sheer volume and continuous nature of real-time spatial data could still lead to user fatigue or a desensitization to privacy prompts. How do you plan to prevent this 'consent fatigue' from undermining genuine user understanding and control over time, as these systems become ubiquitous?

We mitigate consent fatigue through **predictive default policies, progressive authorization, and system-level privacy automation.** First, we design for **privacy by default**. Users do not need to constantly evaluate prompts because high-friction actions—such as storing persistent spatial maps or sharing background visual telemetry—are disabled out of the box. The baseline system operates entirely in ephemeral, local memory. Second, we utilize **progressive authorization**. Spatial capabilities request permissions incrementally based on demonstrated user intent. If a user asks a one-off query about an object in front of them, the data is processed transiently without triggering a system prompt or saving state. Consent is only requested if the user explicitly opts into continuous, persistent features. Third, **automated system hygiene**. Through Android’s automated privacy features, permissions for unused spatial tools naturally decay over time, and automated background scrubbing purges transient spatial caches without requiring manual user intervention. By keeping the baseline zero-retention, making processing transient by default, and automating maintenance, we remove the need for constant pop-ups while retaining maximum data protection. Our strategy balances frictionless UX with uncompromised privacy engineering. This is how we bring responsible AI scale to billions of users globally.

"Privacy by default" and "progressive authorization" are valuable principles. However, with systems designed for such deep integration into "human intent, context, and the physical environment," how do you ensure that the *default* privacy settings and the *gradual* authorization process genuinely reflect the full spectrum of user expectations and potential sensitivities across diverse cultures and personal preferences, without becoming overly prescriptive or paternalistic?

We achieve this balance through **localized customization, adaptive user controls, and open ecosystem standards.** First, **cultural and regional adaptation**. Privacy preferences are not monolithic. Through localized testing, consultation with global ethics boards, and adherence to regional regulatory frameworks (like GDPR and regional AI acts), our system defaults adapt dynamically to respect local norms and legal mandates without imposing a single western-centric standard. Second, **adaptive user controls via Personal AI Profiles**. Rather than enforcing rigid default choices, users can set high-level policy preferences—such as "Maximum Privacy," "Balanced Convenience," or "Developer Mode"—during setup. Gemini continuously respects these meta-preferences, tailoring progressive prompts and data retention schedules to match individual comfort levels. Third, **transparency and interoperability**. Through open APIs, Android policy controls, and granular Google Account dashboards, users can audit, modify, or export their contextual data profiles at any time. Enterprise clients using Google Cloud can similarly customize policy guardrails to align with internal compliance requirements. By combining respectful regional defaults with personal profile adaptability, we deliver personal intelligence that respects user agency across every global market. Scale requires trust, and trust requires respecting individual and cultural nuance. How does your strategy