On Friday, 18 September 2026, CNN published an extraordinary report under the headline: “US military had close call after using AI for false intelligence report, sources say.”
According to CNN, an American military analyst used an artificial-intelligence chatbot while examining intelligence concerning the cargo manifest of a Chinese ship in the Middle East. The chatbot allegedly misidentified the cargo as components connected with a nuclear weapons programme. An intelligence report was then prepared and circulated, military aircraft were put into the air and armed personnel supposedly began preparing to board the vessel.
The operation, we are told, was halted only when officials examined the report more closely and discovered that its conclusion was false.
One unnamed source reportedly described the intelligence as “entirely false”. The same source claimed that it “almost started a war”.
That is an astonishing allegation. It is also one for which CNN presents no independently verifiable evidence.
There is no published intelligence report. There is no identified ship. There is no date for the alleged operation beyond the vague statement that it occurred “this spring”. The cargo supposedly misidentified by the chatbot remains unknown. No military unit involved in the planned interception is identified. No commander has spoken publicly. Neither the Pentagon nor the relevant special operations command confirmed the incident.
Even the particular AI system supposedly responsible cannot be named.
CNN expressly acknowledges that it could not determine whether the chatbot was a commercially available product or an internal government system. Therefore, despite the surrounding political discussion about President Donald Trump’s preferred AI providers, there is no factual basis for attributing this alleged incident to Anthropic, Claude or any other particular company or model.
What remains is a sensational narrative resting entirely upon unnamed sources.
CNN actually refers to four sources as being familiar with the broader episode, although only two are said to support the claim that armed military personnel were preparing to board the ship. One source and another person familiar with the incident reportedly said that military aircraft were already in the air.
That distinction matters, but it does not solve the fundamental problem. Four unnamed sources are still four people whom the reader cannot identify, question or assess. We do not know their positions, whether they witnessed the events directly, whether their accounts are independent of one another or whether they are repeating the same story circulating within government.
We are simply expected to believe them.
Anonymous sources sometimes have a legitimate place in journalism. Intelligence and military personnel may face genuine consequences for speaking publicly. Anonymity, however, does not transform an allegation into established fact. The more extraordinary the claim, the greater the need for corroborating evidence.
CNN is not merely alleging that an analyst made a mistake. It is suggesting that artificial intelligence brought the United States to the edge of war with China. Yet the report supplies nothing that would allow the public to determine whether the incident occurred as described, whether its significance has been exaggerated or whether separate and much more ordinary events have been assembled into a frightening story.
The language does much of the work. An intelligence report “set off alarm bells”. The military “swung into action”. Armed personnel prepared to board. Aircraft were already flying. War was almost started.
Each new sentence raises the tension, but none supplies the missing evidence.
This is especially important because the story arrives at a moment when the public is already being conditioned to fear artificial intelligence. AI is routinely presented as an unknowable force escaping human control, threatening employment, corrupting information and perhaps even endangering civilisation. Once that atmosphere has been created, a story claiming that AI “almost started a war” does not need to be demonstrated. It needs only to be repeated.
And repeated it has been. News organisations around the world have republished CNN’s allegation, often stripping away even the limited qualifications contained in the original report. Within hours, a claim made by unnamed sources becomes an “incident”; the incident becomes an established fact; and the established fact becomes evidence that AI is dangerously uncontrollable.
This is circular journalism in its most effective form. One organisation publishes an allegation. Other outlets report that the allegation has been published. Commentators then cite the resulting volume of coverage as proof that the original allegation must be true.
Repetition is not corroboration.
There is a further question that the anti-AI framing conveniently obscures. Even if every element of CNN’s story is accurate, it would not demonstrate that artificial intelligence nearly started a war. It would demonstrate that a human analyst relied upon an unverified output, placed it into an intelligence report and circulated it without adequate checking. It would then demonstrate that other human beings treated that report as reliable without examining its evidential foundation.
That would be a failure of military intelligence procedures, professional judgement and human supervision.
AI does not hold military rank. It cannot authorise an interception, dispatch an aircraft or order armed personnel to board a ship. It produces material for people to assess. If an analyst copies an unverified assertion into an official report, the analyst has failed. If commanders act upon it without corroboration, the command structure has failed. Blaming the tool is a convenient means of avoiding responsibility for the people and procedures involved.
A calculator can produce a disastrous result when supplied with the wrong figures. A database can contain false information. A satellite image can be misinterpreted. A human source can lie. None of this renders calculators, databases, satellites or human intelligence inherently unusable. It means that consequential decisions require verification from multiple reliable sources.
The principle is no different when AI is involved.
The episode also calls to mind what the author Michael Crichton called the Gell-Mann Amnesia Effect. A person reads a newspaper report concerning a subject he understands well and immediately recognises that the journalist has misunderstood the facts, omitted essential context or even reversed cause and effect. He then turns the page, reads about a subject outside his own expertise and accepts the same newspaper as authoritative.
The lesson discovered on one page is forgotten on the next.
Many readers have repeatedly watched major news organisations publish stories that later proved exaggerated, incomplete or wrong. They know how unnamed officials are used to promote institutional and political narratives. They know that qualifications disappear as allegations travel through the media system. Yet when the subject becomes artificial intelligence—a technology they may use but do not fully understand—the old scepticism vanishes.
CNN itself is hardly an uncontested source of authority. President Trump has spent years describing it as “fake news” and has now announced its exclusion, alongside Politico and MS NOW, from White House access. One need not endorse that decision, or every criticism Trump has made of the press, to recognise the extraordinary political backdrop. CNN is reporting an allegation that can be used to attack the rapid adoption of AI promoted by the very administration with which the network is openly at war.
That does not automatically make the story false. It does make scrutiny indispensable.
For now, CNN has established only that unnamed people told its journalists a dramatic story. It has not demonstrated that an AI system independently invented the crucial claim, that the analyst used the system in the manner described, that an interception was imminent or that war with China was genuinely close.
Perhaps supporting evidence will eventually emerge. Perhaps an official investigation will identify the report, the vessel, the cargo, the AI system and the chain of decisions. If that happens, the evidence can be examined on its merits.
Until then, “AI almost started a war” is not a verified fact. It is a headline constructed around an unprovable allegation and circulated at precisely the moment when fear of AI is politically and commercially useful.
The public should not confuse a frightening story with evidence. Nor should it allow those who wish to restrict artificial intelligence to turn every alleged instance of human carelessness into an indictment of the technology itself.
If CNN wants us to believe that the world came close to war, it must do considerably better than “sources say”.