Bernie Sanders and Claude: The Interview That Misses the Mark on AI

American Senator Bernie Sanders recently released a video on the platform X, dated March 19, 2026, where he engages in a discussion with Claude, the artificial intelligence developed by Anthropic. In this video, Sanders addresses the issue of massive personal data collection by AI systems and advocates for stricter regulation of this rapidly expanding sector. While the intention to protect personal data is commendable, the method employed by Sanders to illustrate his point has notable technical shortcomings.

Sanders is actively campaigning for a moratorium on the construction of new data centers dedicated to artificial intelligence. In his attempt to demonstrate the inherent dangers of data collection, he questions Claude about its practices. However, this approach is misleading because Claude, as a language model, does not have the ability to visualize or understand the internal processes of its operation. It cannot access data pipelines, retention policies, or internal access. Its responses are generated based on probabilities learned from vast corpuses of text, rather than from direct observation of data collection practices.

A legitimate question, but poorly addressed

The questions raised by Sanders are undeniably relevant. They concern what AI collects, how this data is used, and the potential surprises this could hold for the American public. However, the choice of Claude as a conversation partner is problematic. Indeed, Claude, as a language model, does not actively collect data in the way a social network or an advertising tracking system would. Data collection is actually dependent on the services and infrastructure surrounding the model, rather than the model itself.

Confusion between types of data

Sanders' interview with Claude also generates confusion between several categories of data. Notably, there is a distinction between training data, which are the corpuses used to train the model, usage data, which includes prompts, logs, and interactions with users, and data from other digital services, such as browsing, advertising tracking, or online purchases. These different types of data are subject to distinct uses, circuits, and regulatory frameworks. Mixing them in the same discourse prevents a clear understanding of the actual mechanisms. Claude's response is limited to reformulating general knowledge about surveillance capitalism, without providing specific information about Anthropic's practices.

Conclusion

The exchange between Bernie Sanders and Claude highlights more the model's ability to produce a coherent narrative than its ability to inform about its own operating conditions. To obtain a credible assessment of privacy protection, it is essential to conduct independent audits, demand transparency regarding data flows, and carry out investigations based on documentation. Relying on a scripted interview with an AI model cannot replace these essential steps to ensure transparency and the protection of personal data.