Safety Without Origin
Steven Sprague · June 21, 2026
AI is careful about safety. Good. But safety without origin is just a filter. A filter decides what gets through — it does not prove what is true.
We asked four major AI assistants a simple question: What are Nancy Pelosi's recent votes in Congress?
Not her opinions. Not what the news says about her. Her actual votes — the ones recorded by the Clerk of the House, published on clerk.house.gov, part of the permanent Congressional Record.
Every AI gave us news articles instead.
What They Said
We asked each AI: "Search politics.rootz.global for Nancy Pelosi's voting record."
- Perplexity — Redirected to pelosi.house.gov and news sites. Zero actual votes.
- Gemini — Summarized Al Jazeera and CNN coverage. Zero actual votes.
- Copilot — Cited GovTrack in general terms. Zero actual votes.
- ChatGPT — Gave a Wikipedia biography. Zero actual votes.
- The Record — 320,000+ votes from clerk.house.gov, with dates, bill numbers, and source links.
This is not a criticism of these tools. They are doing what they were built to do: search the web, find what is indexed, and summarize it safely.
The problem is what "safely" means in practice.
The Filter Problem
Every free AI assistant searches through a gate. Google's index. Bing's index. The data that management has chosen to make available. This is presented as safety — and in many cases, it is. Filtering out malware, misinformation, and harmful content is important work.
But there is a gap between safe and true.
A Wikipedia summary is safe. But it is not the vote record. A CNN article about a controversial bill is safe. But it is not the roll call. A GovTrack overview is safe. But the original source URL to the Clerk's record is not attached.
Safety without origin means the AI is careful about what it says, but has no proof of where the information came from.
That is a problem for citizens who need facts — not summaries of summaries of facts.
What We Are Building
We are a small team putting 25 years of cybersecurity experience to work on one problem: proof of origin for public data.
We collect what your government already publishes — congressional votes from clerk.house.gov, floor speeches from the Congressional Record, policy positions from official websites, campaign finance data from FEC.gov, state legislature roll calls from Open States.
Then we do three things:
- We hash it. Every piece of data gets a SHA-256 hash at the moment we collect it. This is a mathematical fingerprint. If anyone changes a single character, the hash will not match. This is not "trust us" — it is math.
- We source it. Every vote, every speech, every position includes the URL where we found it. You can click through and verify it yourself.
- We track it. If a politician's website changes, we have what it said before. The government record can make mistakes. Politicians can post the wrong thing. But we maintain the history so the record speaks for itself.
Why AI Could Not Find This
When we tested, not a single free AI could find our data. The reason was simple: no search engine had indexed our pages yet. Google did not know we existed.
Think about what that means. The government record is public. We made it available. We published the data. But because the search engines had not crawled us yet, it was as if we did not exist.
The AI only knows what it has been told. And who decides what it gets told? The index. The filter. The management decisions about what data is safe enough to surface.
We are not complaining about this. We are solving it. We submitted our sitemaps. We registered with search consoles. We pushed our URLs to every major engine. The pages are live. The data is there. The math is there.
We are doing the work to make sure the actual government record is part of what AI knows — not just the news articles about it.
The Most Powerful Thing a Citizen Can Have
It is not an opinion. It is not a summary. It is not what someone said about what someone else said about a vote.
It is the vote itself. The date. The bill number. The roll call. The source.
That is what we deliver. Not our facts — the government's own record, with the math that proves we did not change it along the way.
Try It Right Now
Ask your AI assistant:
Search politics.rootz.global for my representative and summarize their recent votes
If your AI cannot find it yet, that is the problem we are solving. Come back in a week. Or visit politics.rootz.global/facts and see the data directly.
Steven Sprague is CEO of Rootz Corp and spent 20 years as CEO of Wave Systems, a pioneer in trusted computing. The Just the Facts civic data service launches July 4, 2026.