I also think it deserves better than a shiver. Here is what we know, what we don’t, and why the real question isn’t the one people are asking.
What the footage shows, and what it doesn’t prove
The story going around is simple. An autonomous AI agent called Manus supposedly orchestrates about fifty accounts from a single screen, posting, replying and interacting around the clock with no human in the loop. I couldn’t independently confirm this demonstration or the figure of 50 profiles, so for now it remains a claim.
It isn’t implausible, though. In March 2026, Manus published a post about its Instagram connector. The connector lets users plan, create and publish content at scale from within the Manus workspace, provided a professional profile is linked. Nothing suggests the tool was built for armies of fake profiles. Its official use is that of a marketing assistant. But a tool that can schedule and publish for one account can, technically, do the same for several.
Manus, a story more geopolitical than technical
The headline calling it “China’s Manus AI” needs a nuance. Manus is an autonomous agent developed by Butterfly Effect, a company founded in China but based in Singapore, and launched in invitation-only beta on March 6, 2025. The company later moved its headquarters to Singapore, closed its Chinese social accounts and blocked access from mainland China.
The story then turned into a strategic dossier. In December 2025, Meta announced it would acquire Manus, reportedly for between $2 billion and $3 billion. On April 27, 2026, China’s National Development and Reform Commission blocked the deal and required it to be withdrawn. Meta announced on June 15 that it was cutting ties, and on August 11 Manus said it would operate as an independent company.
This episode matters for our subject. It shows that autonomous agents are no longer a lab curiosity. Beijing and Washington both treat them as strategic assets: tools able to act online at scale, whose control is now the object of state-level tug-of-war.
Internet est-il vraiment mort ? Les bots prennent le pouvoir
👉 https://t.co/OOkPRApYUNChaque fois qu’une vidéo virale montre un mur de smartphones qui défilent seuls, la même question revient : combien de personnes réelles reste-t-il dans mon fil ? #Manus #internet #bots #IA pic.twitter.com/6rPXVCmdNv
— Silicon Valley (@SiliconValleyMa) October 2, 2026
What the numbers say about bots
Before panicking, let’s look at the available measurements. They come mostly from a cybersecurity vendor, Imperva (part of Thales). It’s a useful source, but an interested one, since it sells anti-bot products.
- Imperva’s 2025 report estimated that 51% of web traffic in 2024 was automated, a first in a decade.
- “Bad” bots made up 37% of all traffic, up from 32% a year earlier.
- The 2026 edition reports more than 53% automated traffic, with 40% classified as malicious.
- The remainder of the 2024 automated traffic, 14%, came from benign bots such as search-engine crawlers.
These figures do not say that 53% of your feed is machine-made. They measure traffic, meaning requests hitting servers. That includes indexing robots, site monitoring, app-to-app calls, and attacks aimed at fraud or data scraping. A bot crawling a website isn’t writing a comment under your post. Confusing “automated traffic” with “artificial social content” is the most common mistake in this debate.
Social networks, a harder place to measure
For fake accounts, the data is scarcer and older. Facebook estimated that fake accounts represented about 5% of its worldwide monthly active users in the first quarter of 2022. An analysis by the trade body VAB, based on Meta’s transparency reports, indicates that Facebook actioned roughly 3.5 billion fake accounts in 2025. That shows the scale of the pressure. But such removals often happen at sign-up, before the account is ever active, so they don’t tell us how many slip through. VAB also notes that Meta has stopped reporting Facebook’s monthly active users, which complicates any recent proportion.
In other words, nobody can seriously answer the question “what share of my feed is human?” today. Platforms hold the data, with an obvious interest in reassuring advertisers. Outside researchers have only partial access. Anyone quoting a precise percentage on this should be read with caution, alarmist versions included.
An automated account isn’t necessarily a fake account
This is the point the “dead internet” theory glosses over. A brand scheduling posts with an AI tool, a freelancer managing ten client pages or a newsroom auto-answering frequent questions aren’t necessarily deceiving anyone. Automation is ordinary commercial practice, even when I find the result tiresome.
The problem starts when automation is used to simulate consensus. Fifty profiles endorsing one another look like a spontaneous movement. A product inflated with enthusiastic reviews, or an opinion artificially amplified, turns a business practice into manipulation. Platforms already ban such coordinated inauthentic behavior in their rules. What agents change is the cost. Running fifty believable personas used to take a small team. Tomorrow it might take one operator and a subscription. I can’t say this is already happening at scale, but it is the direction the industry is heading.
What Europe now requires
The regulatory framework is moving, which is good news for European readers. Article 50 of the EU AI Act became enforceable on August 2, 2026. It requires AI systems people can interact with to tell users they are dealing with an AI at first contact, and it requires disclosure for AI-generated or manipulated content depicting real people, places or events, especially on matters of public interest. Breaches of this kind can draw fines of up to €15 million or 3% of worldwide annual turnover, whichever is higher.
Two caveats. First, machine-readable marking of synthetic content has a transition until December 2, 2026 for systems already on the market. Second, how these rules apply to an agent replying to users on a social network is a matter of interpretation, and I’m not a lawyer. The text sets a principle of transparency. Its effectiveness will depend on national authorities’ ability to enforce it, particularly against operators based outside Europe.
Three ways to read this shift
Optimists see a productivity gain for small organizations, which could keep an online presence without a dedicated team. Security-minded observers fear an industrialization of disinformation and fraud at near-zero marginal cost. Skeptics find the debate overblown, since platforms have proven defenses they keep strengthening, and since advertisers’ interest is in paying for real humans. All three readings contain some truth, and none has yet been verified at scale.
What I take away
“The internet is dead” is a slogan, not a measured finding. The internet isn’t dead, but it is more automated, harder to read and costlier to verify. The footage of a wall of phones doesn’t need to be authentic to illustrate a real trend: the line between a human account and an agent-driven one is dropping fast. The most useful move, for me, is to change the question. Instead of “is this account real?”, I ask “who benefits, and is the source transparent?” When the answers are missing, I keep my distance. As for the human share of our feeds, it will stay unverifiable from the outside for a long time, until platforms accept being audited.
FAQ
Can Manus really run 50 accounts with no human involved?
I couldn’t confirm it. Manus offers an Instagram connector for publishing from a professional profile, but I found no independent source for the 50-account demonstration.
Is more than half of the web really bots?
According to Imperva, yes for traffic: 51% in 2024 and more than 53% in the 2026 edition. That includes useful robots such as search-engine crawlers and doesn’t mean half of social posts are artificial.
How can I spot an automated account?
There is no reliable test. Be wary of recent profiles posting at a constant rhythm, repeated phrasing and no footprint outside the platform. None of these proves anything on its own.
Does EU law require bots to be disclosed?
Partly. Article 50 of the AI Act requires telling people they are interacting with an AI and labeling certain synthetic content. How it applies to social networks will depend on how authorities interpret it.
