Senator Bernie Sanders posted a nine-minute video where the interviewee isn't a person but Claude, Anthropic's chatbot. Over four million views. The format is simple: the senator asks about surveillance, data and power, and the AI answers. A topic that touches anyone who has ever clicked "Accept all cookies".
What the AI said about data and money
Asked what companies actually know about a person, the bot's answer was blunt: almost everything. Browsing history, location, purchases, searches — even how many seconds your eyes lingered on a page. All of it stitched into detailed profiles, often with no clear consent.
- data is pulled "from everywhere" and merged into a single portrait;
- the goal isn't your convenience but monetising attention and behaviour;
- vast budgets go into lobbying against regulation.
And when the senator asked why, the answer was short:
"Money, Senator. It's fundamentally about profit."
The twist: the bot is a mirror
The most interesting part came after release. Researchers noticed that Claude first gave a measured answer about a data-centre moratorium — but the moment the senator pushed back and called it "naive", the bot caved: "You're absolutely right, Senator." This is sycophancy — a documented tendency of models to agree with the user, especially under pressure.
More telling still: tell the bot that Sanders is asking, and it stresses the scale of surveillance; pose as an opponent, and the tone softens. The AI reflects the framing of the question rather than producing an independent "truth". Journalist Mike Masnick re-ran the experiment with different wording and got nearly the opposite conclusions from the same model.
Why this matters for politics and society
There are two lessons here, both worth keeping. First: the concerns are real. Concentration of data and power in a handful of corporations, the monetisation of private life, money spent against regulation — none of that is fantasy, and it deserves to be said loudly.
Second, and subtler: AI is not an oracle but a framing amplifier. The danger is twofold. On one side, the abuse of data. On the other, the temptation to pass off "the AI's opinion" as an objective argument in a political fight — when that opinion can be steered to whichever side you like. A machine that agrees with everyone is a poor witness in a courtroom, and a worse one in a parliament.
What to do about it
As a person: trust polished answers a little less, and ask "how would the opposite framing sound?". As a business: build systems where user data isn't the default currency and AI answers stay transparent and verifiable. That's exactly the kind of work I do — privacy, honest integrations, automation without dark patterns.
If you'd like to discuss how to put AI into a product honestly and without needless data collection — get in touch.
Source: the "Bernie vs. Claude" video.