Introduction

For years, the debate over democratic technology governance ran on two separate tracks. One track was regulatory: would any government actually enforce rules on the companies building AI? The other was empirical: how fast was the space in which citizens can speak, organise, and dissent online actually shrinking? On August 2, 2026, those two tracks finally converged. That was the day the European Union's AI Act switched on its transparency and general-purpose-AI obligations — even as Freedom House logged a fifteenth consecutive year of global internet-freedom decline.

“For years, the debate over democratic technology governance ran on two separate tracks.”

The timing is not a coincidence. It is a measurement. The EU is betting that enforceable rules can brake a technology that authoritarian states have already weaponised at industrial scale. The data suggests the bet is overdue — and the stakes are enormous.

The Long Slide: Fifteen Years of Falling Internet Freedom

Start with the baseline. Freedom House's Freedom on the Net 2025 report, covering June 2024 to May 2025, found internet freedom declined for the 15th consecutive year. Of the 72 countries assessed — which together account for roughly 89 percent of the world's internet users — conditions deteriorated in 28, while only 17 registered improvements and 27 held roughly steady.

The slide has been relentless and largely one-directional since the project began in 2011. When Freedom on the Net launched, there was genuine optimism that the open internet would spread democratic norms. Instead, authoritarian governments have spent a decade and a half perfecting the tools to dominate digital space — and, critically, to export those tools to one another.

The human cost is now measurable. In 57 of the 72 countries assessed, people were arrested or imprisoned for online expression during the coverage period — a record high for the survey. That is not a fringe phenomenon; it means that in roughly four out of five countries the researchers looked at, saying the wrong thing online carries a real risk of imprisonment.

The pattern is global, but it is sharpest where it started. Countries including Egypt, Pakistan, Russia, Turkey, and Venezuela — which have suffered the most extreme declines in the last 15 years — intensified surveillance of electronic communications and imposed more severe penalties for online dissent, particularly around protests and elections. Kenya recorded the largest single-year decline after authorities responded to nationwide protests with internet blocks and arrests. Even democracies are backsliding: nine of the 18 countries rated "Free" saw their scores fall, with Georgia, Germany, and the United States among the notable decliners.

And the problem is now deeper than censorship. Freedom House reports that 70 percent of the world's population now lives in countries where governments sought to manipulate online content — not merely suppress it. The "online deception" indicator has been the most consistently declining of all 21 measures over the past 15 years, driven by undisclosed pro-government influencers and an accelerating wave of AI-generated content.

The Export Boom: AI Surveillance Has Gone Global

What makes the slide so difficult to reverse is that the tools of digital repression are now a global export industry. The most powerful example is China, which has built the world's largest surveillance apparatus and actively markets it abroad.

The numbers are staggering. China operates an estimated 700 million surveillance cameras at home — more than one for every two citizens — and its tech champions have exported that capability worldwide. Huawei and ZTE have a commercial presence in more than 60 countries. Chinese surveillance technology now reaches 80 or more countries globally, spanning facial-recognition systems, smart-city platforms, and the hardware and software that make state monitoring turnkey.

The economics tell the real story. In March 2026, a study by the UK-based Institute of Development Studies and the African Digital Rights Network found that eleven African countries have collectively spent over $2 billion on AI-powered surveillance systems — much of it purchased from China and financed by Chinese banks. The loans, researchers found, were often conditional on buying Chinese technology and services to build "safe city" systems. This is happening even where most of these countries have little legal regulation or oversight of surveillance.

That is the engine of digital authoritarianism: not a single repressive regime, but a supply chain. And it is precisely the kind of cross-border technology flow that the EU's AI Act — and analogous rulemaking elsewhere — is trying to regulate.

The Regulation: What Actually Changed on August 2, 2026

Amid the gloom, the regulatory picture sharpened dramatically. On June 16, 2026, the European Parliament approved the so-called Digital Omnibus amendments by a vote of 423 to 57, with 174 abstentions. The headline was a delay: most high-risk AI obligations were pushed out 12 to 16 months, to December 2027 and August 2028.

But the delay obscured a more consequential fact. Article 50 transparency duties were not delayed. From August 2, 2026, businesses running customer-facing chatbots for EU users, generating synthetic images or video, or publishing AI-generated content on matters of public interest must comply with rules on: - Chatbot disclosure — telling users they are interacting with an AI system, not a human; - AI-content marking — labelling synthetic or manipulated content; - Deepfake labelling — flagging manipulated images, audio, or video.

The same date activated the European Commission's enforcement powers over general-purpose AI models (GPAI). The maximum fine for violations reaches €15 million or 3 percent of global annual turnover, whichever is higher.

The reason the transparency tier was spared the delay is precisely the topic of this post: these are the tools that directly threaten democratic processes. Deepfake videos of political leaders, AI-generated disinformation, and undisclosed chatbot influence campaigns were already live features of elections in 2024 and 2025. The EU chose to keep guardrails on those even as it gave businesses more time on the thornier high-risk applications like hiring and credit scoring.

The Bottom Line

Two curves are crossing. The curve of internet freedom has been falling for fifteen straight years, and the curve of AI-powered authoritarian surveillance is rising — a $2 billion export boom built on Chinese supply chains reaching 80-plus countries. Against both, the curve of enforcement is only now beginning to rise, with the EU's AI Act turning from paper into law on August 2, 2026.

The lesson for businesses is not abstract. Any company deploying AI in the EU market now has concrete, enforceable transparency obligations — with fines that can reach €15 million or 3 percent of turnover. And any company buying or building AI systems should think hard about where its technology comes from, because the same models that power productivity gains are being repurposed at scale to monitor and repress citizens. Governance is no longer a compliance checkbox. In 2026, it is the front line.

Sources

  • Freedom House, Freedom on the Net 2025: An Uncertain Future for the Global Internet (published Nov 2025) — 72 countries assessed; 15th consecutive year of decline; 28 deteriorated / 17 improved / 27 stable; 57 of 72 countries saw arrests or imprisonment for online expression (record high); 70% of world population under online content manipulation; 9 of 18 "Free" countries declined; Kenya largest decline, Bangladesh strongest improvement; Iceland freest, China and Myanmar most repressive.
  • Freedom House, Freedom in the World 2026 (Mar 2026) — global freedom declined for the 20th consecutive year; 54 countries saw declines, 35 improvements.
  • European Parliament Digital Omnibus amendments (final vote Jun 16, 2026: 423–57, 174 abstentions) — high-risk AI obligations delayed to Dec 2, 2027 / Aug 2028; Article 50 transparency + GPAI enforcement effective Aug 2, 2026; max fine €15M or 3% of global turnover.
  • Institute of Development Studies / African Digital Rights Network (Mar 2026) — eleven African countries spent over $2 billion on AI-powered surveillance systems; Chinese loans conditional on purchasing Chinese tech; most lack legal oversight.
  • Rest of World (Mar 20, 2026) — reporting on the IDS/African Digital Rights Network surveillance study.
  • State of Surveillance / Modern Diplomacy (2026) — China exports surveillance tech to 80+ countries; Huawei/ZTE presence in 63+ countries.
  • Carnegie Endowment, "China's AI-Empowered Censorship" (Mar 10, 2026) — how AI systems now underpin Chinese public and private censorship.
  • Wikipedia, Mass surveillance in China — 700+ million surveillance cameras (2023 estimate).