The truth infrastructure of the internet
Google withdrew its AI image generation feature from Google Earth within 24 hours of launch because it produced fake images of nuclear strikes at real, politically sensitive coordinates. This is a concrete example of how quickly our digital ground truth can be corrupted. When citizens or public offices rely on maps and public records, they do so because they assume a baseline of authenticity that no amount of watermarking can fully restore once it has been damaged.
In fact, the contamination of our information sources is moving faster than the defenses we have in place. Amazon’s Kindle Direct Publishing saw its monthly ebook releases triple to 300,000 by the end of 2025, with most of that growth attributable to AI-generated content. As well, a Nature investigation found tens of thousands of scientific publications from this year alone containing hallucinated references. If the sources we use for decision-making are being flooded with synthetic noise, the risk to public governance is not just a theoretical concern.
I have to admit that building trust in digital systems is already hard enough in the public sector. If we allow AI-generated hallucinations to become part of our shared data, we create a feedback loop where bad science begets worse science. We see this happening. Wikipedia traffic falls as search engines provide direct answers based on a shrinking pool of human-vetted content.
Now we need to treat truth as an INFRASTRUCTURE challenge rather than just a content problem. This requires more than just filters. We need technical standards like C2PA and the specific enforcement of Article 50 of the EU AI Act to ensure that machine-readable content is actually verified. This ensures that verified information remains discoverable. We must protect the integrity of our common data foundations before they become unusable for public service.
#AI #PublicAdministration #DigitalGovernance #DataIntegrity #EUAIAct