EU AI Act transparency phase starts, but with clear gaps

"A screenshot, format conversion, or simple file re-save suffice to break the chain of provenance." That is Taras Kovalchuk's terse read on the state of watermarking, delivered in an op-ed for Tech Policy Press weighing whether the EU AI Act's transparency rules will actually help European democracies.

The rules he is weighing are real and, as of August 2, 2026, enforceable. Article 50 requires that "providers of certain AI systems must ensure that people are informed when they interact directly with an AI system," and that generated content be marked in a machine-readable format. The AI Office picked up full enforcement authority over general-purpose model providers. Staff at deployers must now show basic AI literacy under Article 4.

Kovalchuk's argument is less with the rules than with the physics around them. "Watermarking only addresses the initial generation phase," he writes; the mark "remains vulnerable throughout a much longer lifecycle spanning distribution, modification, detection, and human interpretation."

Then there is who is generating the content in the first place. "Much of the disinformation targeting European democracies originates from Russia, and China," he writes. "The foreign states are beyond the AI Act's jurisdiction and thus cannot be deterred directly." Modified open-weight models running on private infrastructure escape oversight for the same reason.

Telegram gets a specific mention. Per Kovalchuk, the messenger has stayed "just below that threshold for over two years despite EU scrutiny," dodging the DSA's heightened obligations around detecting and mitigating coordinated manipulation.

His close is deliberately narrow. The AI Act "can make synthetic content easier to identify. Whether that makes European democracies more resilient will depend on what governments, platforms, institutions, journalists, and citizens are able to do with that information."