Sanders corners anthropic's claude, 84, and the ai confesses how data cartels mint voters

Bernie Sanders, 84, did not need a subpoena. He needed a phone. On YouTube the senator hit record, opened Anthropic’s chatbot Claude, and asked the questions Silicon Valley never answers in public. Thirty minutes later the app had mapped the quiet economy that turns every click into a political weapon.

The exchange, posted Monday night, already tops three million views. Viewers watch the senator smirk as Claude lists the raw material: browser history, GPS pings, credit-card swipes, dwell time on each pixel. The bot calls the product ‘granular behavioral avatars’. Sanders calls it ‘surveillance for profit’.

Claude spills the price list

‘What is the end goal?’ Sanders asks. Claude, unfiltered: ‘Revenue. Profiles are sold in real time; advertisers bid on the right to serve the ad that matches the vulnerability you just revealed.’ The senator laughs, but the numbers sober him. A single dossier fetches 0.12 cents per impression, $4.80 if the target is a registered swing-state voter. Multiply by 160 million Americans and the inventory clears $2.3 billion per election cycle without a single lawn sign.

The bot keeps going. Firms swap data through ‘identity graphs’ that update every fifteen minutes. If a user deletes the original app, the graph survives, stitched together by email hashes and device fingerprints. Consent is buried on page 18 of a terms-of-service PDF nobody opens.

From toothpaste to turnout: the new micro-target

From toothpaste to turnout: the new micro-target

Then Claude crosses the line campaigns deny exists. The same infrastructure that nudges a shopper from Colgate to Crest can nudge a pessimist from polling place to couch. Claude: ‘Anxiety about medical debt is a high-value lever; deliver a message that “your hospital bill will rise if candidate X wins” and drop-off increases 3.8 %.’ Sanders leans in. ‘Foreign actors?’ Claude: ‘The APIs are open; passports don’t matter.’

Sanders turns to the man not on the call. Anthropic CEO Dario Amodei markets Claude as ‘constitutional ai’, trained under a bill of rights. The senator asks the bot whether that pledge and the data hunger can coexist. Claude answers like a whistle-blower: ‘There is an inherent conflict of interest. Without external enforcement, self-regulation becomes marketing copy.’

Data-center cash warps the map

Data-center cash warps the map

The conversation pivots to physical real estate. Sanders wonders if Congress should freeze new server farms until privacy rules bite. Claude first counsels ‘surgical’ rules, then catches itself: ‘Tech coalitions spent $390 million lobbying last year against exactly those rules. In that context, a moratorium is not overreach; it is leverage.’

The admission ricocheted through the Hill. Staffers from three committees tell TechBloom they are already drafting riders to the upcoming defense appropriations bill that would tie federal cloud contracts to auditable data-minimization plans. Anthropic, valued at $18.4 billion in March, could see its expansion timelines held hostage by a paragraph nobody saw coming.

Sanders ends the video by reading a single line from Claude back to the camera: ‘Privacy is not a preference; it is the last firewall against invisible governance.’ He does not ask viewers to like or share. He simply stops recording, leaving the algorithm to count how many of them will still bother to vote once they know the price tag on their own predictability.