This document provides a technical roadmap describing how Web technologies intersect with emerging AI capabilities. It outlines current work across W3C groups, identifies architectural considerations, and highlights areas where coordination or new technical exploration may be needed. The roadmap is not a specification. Rather, it aims to support discussion, promote shared understanding, and guide the future evolution of the Web in the context of growing AI integration.
This document is still at an early stage, and help and feedback are very welcome. If you think something should be added or adjusted, please feel free to open an issue or submit a pull request.
This document outlines an early roadmap for understanding how AI technologies intersect with the Web platform. Its goals are to identify areas where coordinated technical exploration could benefit the ecosystem, highlight emerging patterns, and support broader architectural discussions across W3C.
This roadmap is intended as a living document and may evolve as the technology landscape develops.
These technologies provide the fundamental infrastructure required to execute machine learning tasks directly within the browser, utilizing underlying hardware for maximum efficiency.
Web Neural Network API (WebNN): This is a dedicated low-level API for neural network inference. It allows the browser to communicate directly with hardware accelerators like GPUs, CPUs, or dedicated NPUs (Neural Processing Units). In the context of Web AI, WebNN is the bridge that enables high-performance, low-latency execution of machine learning models locally on a user's device.
WebGPU: WebGPU is the successor to WebGL, providing modern features for high-performance graphics and data-parallel computation. For Web AI, it is essential because most modern AI frameworks (like TensorFlow.js or Transformers.js) use WebGPU to handle the massive parallel processing required for training and running large models in the browser.
WebAssembly Core Specification (WASM): WebAssembly allows code written in languages like C++ or Rust to run at near-native speeds on the web. It enables developers to port heavy machine learning libraries and complex mathematical kernels to the browser without the performance overhead of traditional JavaScript.
WebRTC: Real-Time Communication in Browsers: This standard enables real-time communication of audio, video, and data. Its relationship with Web AI is primarily found in real-time media processing, such as background blur in video calls, noise cancellation, or real-time transcription, all of which often happen on-device via the browser.
add a table to show the status of each specification.
AI systems rely on the organization and expression of data. W3C’s long-standing work in the Semantic Web provides the structured framework necessary for AI to interpret information accurately.
RDF:
JSON-LD: This is a lightweight format for linked data using JSON. It makes web data machine-readable. In the Web AI ecosystem, JSON-LD is crucial for Semantic SEO and helping AI agents parse and understand the context of a webpage's content without needing complex scraping or unstructured text analysis.
add a table to show the status of each specification.
Software agents capable of performing tasks on behalf of users are receiving increasing attention. These agents may navigate Web content, invoke APIs, or interact with services in semi-autonomous ways. The Web architecture provides a natural environment for such systems, but also introduces new challenges related to permissions, capabilities, user control, and predictable behavior. This subsection outlines work areas and technologies relevant to agent interactions in Web contexts.
WebDriver: This is a remote control interface that enables introspection and control of user agents (browsers). For Web AI, WebDriver is often the tool used to train AI agents to navigate websites, automate form-filling, or perform "browser-use" tasks where the AI acts as a human would within the interface.
WoT:The Web of Things aims to counter fragmentation in IoT. Its connection to Web AI lies in providing a standardized way for AI agents to interact with physical devices (like smart home sensors) via standard web protocols, essentially giving the AI hands in the physical world.
add a table to show the status of each specification.
Decentralized Identifiers (DIDs): Decentralized Identifiers are a new type of identifier that enables verifiable, decentralized digital identity. In Web AI, DIDs help users maintain a persistent and secure identity that is not controlled by a single central authority, which is vital when interacting with AI agents that may need to verify a user's identity.
Verifiable Credentials Data Model: Verifiable Credentials provide a mechanism to express credentials on the Web in a manner that is cryptographically secure and privacy-preserving. This relates to Web AI by allowing users to prove things about themselves (like age or certifications) to an AI service without sharing unnecessary personal data.
Federated Credential Management API: Federated Credential Management is an API for privacy-preserving identity federation. As Web AI becomes more personalized, FedCM allows for seamless login and identity verification while protecting users from the tracking risks often associated with third-party identity providers.
add a table to show the status of each specification.
Privacy Principles: These are high-level guidelines for web technology development. In the context of Web AI, these principles ensure that features like local model execution do not inadvertently create new ways to "fingerprint" a user or leak sensitive data from the browser's memory.
AI Threat modeling: This is a specific exploration of security risks associated with running AI in the browser. It helps developers understand the unique risks of Web AI, such as prompt injection at the browser level or side-channel attacks that could leak information about the model or the data it is processing
add a table to show the status of each specification.
Ethical Principles for Web Machine Learning: This document outlines how to build ML features responsibly. For Web AI, it focuses on issues like fairness, accountability, and transparency, ensuring that browser-based AI doesn't reinforce societal biases or perform discriminatory profiling.
Web User Agents: This explains the role of software that acts on behalf of a user. It is critical to define where the AI assistant ends and the user agent begins, ensuring the user remains in control of their digital experience rather than being directed by an opaque AI
AI & the Web: Understanding and managing the impact of Machine Learning models on the Web: This document proposes an analysis of the systemic impact of AI systems, and in particular ones based on Machine Learning models, on the Web, and the role that Web standardization may play in managing that impact.
The Agentic Web: risks and opportunities: This document proposes an analysis of the potential impact of the deployment of AI agents on the Web, the role that web standardization may play in managing that impact, and how to provide users of these tools with mission-aligned guardrails.
add a table to show the status of each specification.
add a table to show the status of each specification.