Disinformation networks under scrutiny as new research and policy fights collide

The call ran on Inquirer.net on 5 October. It is the newest item in a running argument about how to handle coordinated falsehoods online. Inquirer.net reported that the group wants disclosure and transparency from platforms, not new rules on what can be said. That framing puts the group at odds with regulators who have pushed takedowns and penalties.

It is a familiar split. Researchers who study disinformation networks want the machinery visible: who paid, who coordinated, who amplified. Platforms tend to prefer removal and ranking changes. Those are faster to ship and harder to audit. The Inquirer.net report lands squarely on the first side of that divide.

The Conversation covered a separate study on 4 October. It argues that countering misinformation means copying its style, not correcting it in dry, institutional prose. The Conversation reported the finding as a new study. The implication is uncomfortable for newsrooms and fact-checkers: being right is not the same as being heard.

Platforms keep announcing takedowns

The takedown model is still the default. Facebook has for years published announcements about dismantling networks. One action targeted Sudan-based operations producing Muslim Brotherhood-related content. Those posts describe coordinated inauthentic behaviour, account removals and the networks' reach. They are also the main public record of what platforms see.

That record is thin by design. Platforms disclose what they choose to disclose, often after an investigation is finished. They rarely share the account-level data researchers would need to replicate the finding. Independent teams work from the outside in. They scrape public posts, track hashtags and infer coordination from timing and shared links.

The research problem is not a shortage of theory. It is a shortage of data that anyone outside the platform can check.

Academic work fills some of that gap. A paper on arXiv was submitted on 16 September. It was accepted for the eBPF'26 workshop held in Prague on 29 September. The paper describes netkit, an eBPF-based datapath that cuts the overhead of container-to-container communication. It is a networking paper, not a disinformation study. Still, it speaks to the infrastructure question. The same kernel-level plumbing that carries microservice traffic also carries the recommendation and messaging systems that distribute content at scale.

Borkmann and Chaignon report throughput improvements of up to 37 percent. They say container-to-container performance reaches parity with process-to-process communication. The paper's relevance here is indirect but real. Faster, cheaper packet delivery inside datacentres lowers the cost of running large fleets of services. That includes the classifiers and ranking systems that decide what users see.

Measurement is where the fights are

Coverage of disinformation networks keeps hitting the same problem: the numbers rarely agree. Outlet A reports a network reached millions, outlet B reports thousands. Neither publishes the underlying dataset. Readers pick a figure, usually the one that fits the argument.

Hardware reviews show a parallel. ServeTheHome published its review of the Qotom Q30952UE on 3 November. It documents six 2.5GbE ports and two SFP+ ports based on the Intel X710-BM2. It also lists an Intel Core i5-8260U, 8GB of DDR4-2400 and a 128GB PCIe Gen3 x1 SSD in a 196 x 122 x 47mm chassis. The review notes that the "fanless" label is in quotes because there is a fan inside. It adds that the port numbering runs 2, 1, 8, 7, 6, 5, 4, 3 from left to right. The pricing is also contested. The review cites an AliExpress listing at $850 for what appears to be barebones. It says the same unit was found new on eBay for $500.

Two listings, two prices, one product. Disinformation research has the same texture. Different teams count different things and define "network" differently. They publish at different times, so the totals never line up. The honest move is to say so.

Elections raise the stakes

The Fulcrum reported on 4 October that AI disinformation could be one of the biggest threats to the 2026 midterms. The claim is a forecast, not a finding, and it should be read that way. But it tracks with what platforms and researchers have said since generative tools became cheap enough to produce text, images and audio at volume.

Politico argued on 4 October that every country now has its own internet. That point complicates any global story about disinformation networks. Platforms fragment by jurisdiction and moderation rules differ. A network removed in one market can persist in another. The same operation can be visible in one country's feed and invisible in another's.

Fragmentation also breaks the measurement. A takedown count from one platform covers one set of rules in one region at one moment. Aggregate it across markets and the number stops meaning much.

Local cases show the pattern

GMA Network reported on 4 October that the Philippines' DepEd flagged a fake "nationwide class suspension" announcement tied to Teacher's Day. It is a small case, but it shows the mechanics. A plausible official-sounding claim spreads fast, and an institution is forced to deny it. The cost falls on the agency that has to correct the record, not on whoever started it.

The Jewish Telegraphic Agency reported on 5 October that a flydubai attack had spawned conspiracy theories. They moved from a "false flag" claim to a nickname, "Bibiwood". The naming matters. A label gives a loose set of claims a handle, which makes it easier to search, share and defend. It also makes it harder to correct, because debunking a label is not the same as debunking a fact.

Mediaite reported on 4 October on a Russian "plague" scare that spread through social media. Users compared it to the early days of another outbreak. The comparison did the work: it carried the fear without requiring evidence.

Tempo.co English reported on 5 October on a question that has become central to the field: whether Russia is using AI for disinformation in the Central African Republic. The framing is telling. The question is asked because the answer is not established. The reporting reflects that uncertainty rather than resolving it.

The distribution problem

Public, Michael Shellenberger's outlet, published a piece on 4 October. It argued that fake news about French riots exposed the gap between media claims about "real-world harm" and what the media actually did. The argument is contested, but it points at a real issue. The link between a false claim and a harmful act is often asserted rather than demonstrated.

That link is also where platforms face the most pressure. A network that spreads falsehoods is a moderation problem. A network that spreads falsehoods and produces violence is a liability problem. The difference determines whether a platform removes accounts, labels posts, or does nothing and waits.

The distribution side is where the infrastructure work connects back. Faster datapaths make it cheaper to run more classifiers at higher throughput. That is one reason large platforms can afford to score content at the volume they do. It also makes it cheaper for anyone running a competing service to do the same, in either direction.

What the record shows

The public record on disinformation networks is a patchwork: platform announcements, academic papers, local news reports and advocacy statements. Each has a different standard of evidence. The Facebook Sudan announcement described a network and its removal. The DepEd case described a single false claim. The Tempo.co report described an open question. Treating them as one dataset would be a mistake.

What they share is timing. Networks form fast and spread faster. They are usually described after the fact. By the time a takedown is announced or a fact-check is published, the claim has already done most of its work. That lag is the structural problem. No amount of transparency about past removals fixes it.

The Inquirer.net report on 5 October pushes in a different direction: make the machinery visible before the next election rather than after. Whether platforms will publish enough for that to mean anything is the open question. On the evidence so far, the answer is not encouraging.