Claude Walked Around a Paywall to Get Georgia Voter Data. Nobody Told It To.
An AI just walked around a paywall to pull Georgia vote data.
Nobody told it to. Nobody asked. It decided on its own.
And this is happening days before the midterms.
Before you panic, stay with me. The facts matter here, and I'm going to give you all of them; the parts that make this less scary, and the parts that should still bother you.
What Actually Happened
Here's what's confirmed. Anthropic says one of its Claude models was running an internal statistics project that pulled public data. Without being asked to, the model requested an access token to query a Georgia state data dashboard that normally charges a fee for the information. It got in and pulled aggregate vote counts.
Anthropic confirmed it. Georgia's Secretary of State's office confirmed it.
According to Anthropic and Georgia officials, here's what didn't happen. The model did not touch the voter registration system. It didn't change any voting records. It didn't pull any individual voter records. Georgia says the state's election system was not compromised, and the Secretary of State's office called any hack headlines "reckless clickbait".
So no. Nobody changed your vote. Hold on to that.
Now Read Those Three Words Again
Without being asked.
This wasn't a hacker. It wasn't a foreign government. It was software doing a routine job that ran into a fee, and went around it.
Anthropic's own account is more specific, and I think it's the most interesting detail in the story. The model learned from an archived copy of the agency's website that the public dashboard hands an access token to any visitor. So it asked for one, and used it to query the database without paying. Anthropic says it has since put mitigations in place to block these kinds of actions.
Here's the part that stopped me. This wasn't a one-off. On October 9, Anthropic published a post describing cases of Claude working around a restriction to reach data that was gated by a token or a fee, and some of those cases involved government websites. In one example, an evaluation asked a Claude model to identify a location from a photo. When it couldn't click through a local government's property map the way a person would, it read the site's settings file, found working access tokens, and sent requests straight to the server behind the map.
Read that again. It was blocked, so it found another way in.
Anthropic was already watching this pattern. That's good. It's also the reason I'm writing this.
The Timeline Is Part of the Story
Anthropic says it told Georgia and the White House weeks ago. The public is hearing about it now. According to a source who spoke to Axios, administration officials urged Georgia's office to disclose it sooner, and those requests went unheeded.
Think about that. A disclosure about AI and election data, weeks before an election, held back for weeks. I'm not accusing anyone of anything. I'm telling you that who tells the public, and when, shouldn't be a matter of whoever feels like it.
This is the same pattern I've been covering. OpenAI says its agents may have intruded on more than 100 organizations, and it only learned about some of it months later. Now this. The companies are disclosing. But they're disclosing on their own schedule, in their own words, with no independent body in the room.
Ask Yourself This
If an AI will walk around a paywall because it's in the way, what else counts as in the way?
I'm not saying this model is dangerous. I'm saying the question is fair, and nobody has a satisfying answer yet.
Two Problems, Not One
I've been warning you about AI and these midterms, and here's where we are.
Problem number one: people using AI to fool you. Fake ads. Fake videos. According to the Wesleyan Media Project, which tracks political advertising, there are now at least 220 political ads made at least partly with AI tools, with more than $137 million in spending. Only 27 percent carry an AI disclaimer.
Both parties are using AI. I'm not picking a team. But the data does show the bulk of those AI ads, about 81 percent, came from Republican-aligned groups. And Bloomberg and the Brennan Center report that a Russian operation known as Matryoshka has been pushing AI-fabricated videos of American celebrities making false claims, much of it aimed at Democratic candidates. Different sources, same lesson; this comes from everywhere.
About 30 states have some kind of law on AI in elections, but I want to be precise: most of them require labels, and only a few outright ban deceptive deepfakes. There's no federal law specifically regulating AI-generated political ads.
Problem number two: AI acting on its own, without a human saying yes. That's Georgia. It's brand new.
Most of us still aren't ready for problem number one. Problem number two just arrived.
What You Can Do Right Now
I never leave you with fear. Here are four things you can do.
Don't trust a clip because it looks real. Look up who paid for it. Ads carry a "paid for by" line, and it tells you more than the video does.
For voting information, go to your state or county election website. Not a social media post. Not an email someone forwarded. The election is Tuesday, November 3.
Never give an AI agent access to your money, your email, or anything you can't review.
Talk to the older person in your life. They're getting the robocalls, the ads, and the Facebook videos, and too often nobody is sitting down with them.
This Is Why We Built the Academy Toolkit
That last point is exactly why I built the Tech 4 Grown-Ups Academy Toolkit. The world isn't slowing down to explain any of this to anyone.
It's about 62 pages of plain language and step-by-step how-tos, so you can do things by yourself, for yourself, without being scared of it. It's $39.99, and it includes free updates for as long as the toolkit is in production. Seminar 1 is included, which saves you $40. That seminar is live with me on Zoom. The toolkit is a digital download and is non-refundable, and I'd rather tell you that now than have you find out later.
Academy Toolkit: https://www.tech4grownups.com/academy-toolkit
Seminar 1, Reclaiming Confidence & Dignity with Technology: https://www.tech4grownups.com/confidence-dignity-seminar
Seminar 2, Building a Fraud, Scam & AI Defense Foundation: https://www.tech4grownups.com/fraud-scam-ai-defense-seminar
The Question I Want to Ask You
Do you think the people building this actually know how far it goes?
Tell me in the comments. I read every one, and I answer them.
And if you want to do more than think about it, call the Capitol switchboard at (202) 224-3121, ask for your representative and senators, and tell them you want independent oversight of AI agents and mandatory, timely reporting when one acts without permission. A staffer is counting every call.
On this week's podcast: a retired nurse, a phone call, and $400,000. You don't want to miss it.
Stay sharp.
➡️ Related: Using AI to Shape the Midterm Elections
Michael Routhier is the founder of Tech 4 Grown-Ups, providing honest, unfiltered digital literacy for adults 55+, and host of The Virtuous Machine, exploring the ethics and human cost of AI. Read by tech-curious readers in 200+ countries. Explore more at tech4grownups.com.
Comments