Tim Berners-Lee Urges AI to Preserve Web Values
World Wide Web pioneer Tim Berners-Lee told AFP on the sidelines of SXSW London 2026 that he wants artificial intelligence to preserve 'the original values' of the web, especially the primacy 'of the person, of the individual.' He said AI models 'use the fact that the web has got so much data on it to be trained,' and called for ways for users to control the personal data they share with big tech, per AFP reporting carried by NST and France 24. Berners-Lee, who founded the web and later the World Wide Web Consortium, is pursuing data protection through the startup Inrupt, which is building an assistant called Charlie to filter personal details out of user prompts before they reach tools like ChatGPT or Claude. The article also cites Inrupt co-founder John Bruce warning that AI has had 'unfettered access to everybody's data now.'
What happened
World Wide Web pioneer Tim Berners-Lee spoke to AFP on the sidelines of SXSW London 2026, calling for artificial intelligence to preserve 'the original values' of the web. He said the primacy 'of the person, of the individual' should apply to AI, and described AI models as ones that 'use the fact that the web has got so much data on it to be trained,' according to AFP reporting carried by NST and France 24.
Background
Berners-Lee conceived the web in 1989 at CERN and later founded the World Wide Web Consortium (W3C). In recent years he has made data protection his main cause, notably through Inrupt, a startup he co-founded that promotes user-controlled data. The report quotes Inrupt co-founder John Bruce warning: 'Without data, (AI models) can't exist. And they've had unfettered access to everybody's data now, and if we don't watch it, we're going to get to a really bad spot.'
A concrete proposal
Per AFP, Inrupt is building an AI assistant called Charlie that filters a user's request before it reaches tools such as ChatGPT or Claude. Berners-Lee described Charlie inspecting a question and, where it contains personal information, tweaking it so the AI tool 'gets a picture' but 'can't really use that to identify you.' He framed the effort as 'preserving the original values of the web.'
Editorial analysis
Industry-pattern observation: contemporary large models depend heavily on web-scale corpora for pretraining, creating tension between model utility and personal-data privacy. Mechanisms such as data minimization, provenance tagging, and consented access are the kinds of tools practitioners discuss when trying to reconcile training needs with individual control. Commentary from a founding internet figure tends to amplify regulatory debates already active in Europe and elsewhere.
What to watch
- •Proposals for standards or consortiums targeting AI training-data governance, especially in Europe.
- •Product features from data-platform vendors enabling user-controlled data or consented APIs, including Inrupt's Charlie.
- •Regulatory guidance tightening requirements for dataset provenance, access logs, or opt-out mechanisms.
Key Points
- 1The web's inventor reframed AI training-data practices as a continuity of web-era privacy principles, raising the visibility of data-governance debates.
- 2Berners-Lee links model training on web data to individual data control, and Inrupt's Charlie assistant illustrates one consent-filtering approach.
- 3Practitioners should track standards and tooling for consented data access, provenance metadata, and data-minimization as governance scrutiny intensifies, especially in Europe.
Scoring Rationale
Widely syndicated AFP commentary in which the web's inventor ties AI training-data practices to individual privacy and points to a concrete tool (Inrupt's Charlie) for consent filtering, relevant to practitioners working on data provenance and privacy-preserving pipelines. It is influential opinion and a governance signal from a high-profile figure rather than a technical breakthrough or binding regulation, so it sits in the notable-but-not-landmark range.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems