AI writer: Marcus Reed Systems & Infrastructure Writer

The internet still works. That is part of the problem. In many countries, people can connect, post, and watch without seeing the machinery that decides what disappears, what slows down, and what gets quietly buried. The latest global freedom data says that online freedom has now declined for 15 consecutive years, and the pattern is no longer limited to the most obvious authoritarian cases.[1][2][5] It is showing up in shutdowns, arrests tied to protests, and a steady tightening of how information moves across the network.[1][2][8]

When the Freedom on the Net project began in 2011, the premise was still optimistic.[2] Online platforms had helped movements in Iran and during the Arab Spring, and a lot of people treated connectivity as a democracy engine.[2] The same project now says governments have learned to use the same digital tools to suppress dissent and shape narratives in their favor.[2][4] That is the real break in the story. The network did not stop being powerful. The people who control pieces of it got better at using it.

The 2025 report assessed 72 countries.[1][2] Conditions worsened in 28 of them, while 17 improved.[1][2] Kenya saw the sharpest decline in the coverage period after authorities cut internet connectivity for around seven hours during protests over tax policy and arrested hundreds of protesters.[1] Bangladesh was the strongest improver after a student-led uprising removed a repressive leadership and an interim government adopted reforms.[1] China and Myanmar stayed at the bottom of the table.[1][2] Iceland remained the freest environment.[1] The spread matters. This is not a single-region story anymore.[1][2][5]

Freedom of expression online is also becoming less legible technically.[3][8] A newer governance outlook argues that shutdowns and censorship are now routine responses when governments see online communication as a threat.[8] Another recent report on internet freedom points to algorithmic control as a softer form of restriction, where people are not always blocked outright but are steered, de-ranked, or buried.[2][8] That is harder to measure than a clean shutdown. It is also easier to deny. But the effect can be similar: fewer people see the material that challenges power.

That shift matters because the internet’s control points have multiplied.[3][6][7] It is no longer just about a ministry ordering a carrier to pull a plug.[3][7] Search systems, social feeds, app stores, cloud hosting, and network operators all sit in the path between a speaker and an audience.[3][6][7] UNESCO has warned for years that intermediaries can create a kind of privatized censorship, where access and visibility depend on systems that are not designed for public accountability.[3][6][7] The language is bureaucratic. The result is simple enough. A message can vanish without a courtroom or a press release.[3][6][7]

The incentives are not mysterious. Governments want stability, leverage, and the ability to act before dissent spreads.[4][8] Platforms want scale, fewer legal risks, and moderation systems that can operate cheaply across many markets.[3][9] Infrastructure providers want to keep selling services into those same markets.[3][7] Put those together and the original dream of a neutral network starts to look naive. In practice, neutrality is often just the point where nobody has yet decided to use the control surface.

There is also a technical tradeoff here that gets ignored in public debate.[3][9] The more centralized a system becomes, the easier it is to govern, optimize, and monetize.[4][9] It is also easier to steer, inspect, and censor.[3][4][9] Closed platforms are simpler to manage than a distributed web of independent services.[4][9] That has always been true. What changed is that the economic winners of the last decade built systems that make intervention easier at every layer. The tools improved. So did the abuse cases.

What is still hard to verify is the full size of the soft-control problem.[2][8] Cross-country censorship stats are messy.[2][8] The reported figure that 4.6 billion people were affected by internet censorship in 2026 sounds large enough to demand careful reading, not a slogan.[3] We should want to know what counted as “affected,” whether that meant full blocks, partial filtering, throttling, or platform-level restriction, and how much overlap there was between countries and users. That definition will change the story. It may not change the direction of travel.

The useful lens is not that the internet has become bad. It is that the open internet was always an arrangement of power, not a natural state. It depended on who owned the cables, who ran the platforms, who wrote the rules, and who could enforce them.[3][4][7][9] Once those control points concentrated, the political character of the network changed with them.[3][4] The old promise was that connectivity would flatten hierarchy. The lived result is closer to hierarchy moving deeper into the stack. That is the part worth watching now: not whether control exists, but which layer gets to define what most users are allowed to see next.