Edition 14 · September 25, 2026
According to reporting repeated across several outlets, a US special-operations analyst used an AI chatbot to help assess cargo aboard a Chinese ship. The resulting intelligence report wrongly identified the cargo as components for a nuclear-weapons programme. It travelled far enough through the military system for aircraft and armed personnel reportedly to prepare to intercept the vessel.
The obvious lesson is that chatbots invent things. We knew that already. The more important question is how an invented claim acquired the authority of military intelligence. Was this an early example of almost catastrophic cognitive offloading?
The claim became harder and harder to question as it acquired more credibility the further it travelled. While a chatbot produced an answer, a human analyst ultimately turned it into a report which then entered official channels and triggered a cascading failure of trust. The people receiving it could reasonably assume that somebody upstream should have checked the evidence. And so a weak claim became an institutional fact simply because the process around it made it look authoritative.
There were humans in the loop throughout. But “Human oversight” means little when responsibility is distributed across a chain of people who each assume that verification happened somewhere else. What matters is whether the raw evidence remains visible, whether claims can be traced back to their sources, and whether someone is explicitly responsible for challenging them before action follows. We may need to invent a new role for an AI Devils Advocate.
The public debate about AI and disinformation usually concerns fake videos, automated personas and propaganda flooding social media. But this incident points to a more serious route. False information can enter closed, trusted systems masquerading as vetted professional analysis. It does not need to go viral. It only needs to survive one bureaucratic oversight.
The account itself still requires a bit of caution. While Aunoo found it repeated widely, most of the coverage appears to derive from the same underlying single report. Repetition is not corroboration of truth. Even that is part of the lesson.
Homo sapiens means “the wise ape”, and I have read us described elsewhere as homo ludus, “the playing ape” , but I often feel homo fallax, the deceptive ape, would be the better description
And so the most consequential AI-generated falsehood of the week was not some culture war TikTok deepfake. Instead it may have been a report about people with weapons implicitly trusting AI slop.
The report that nearly sent US forces after the wrong ship
- A US special-operations analyst used an AI chatbot to help produce a report that wrongly identified cargo aboard a Chinese ship as nuclear-weapons components (PCMag). The report circulated through military channels and nearly led US forces to intercept or board the vessel.
- The assessment was later described as “entirely false” (Mashable). Numerous outlets have repeated the story, although most appear to draw on the same original account. There is still no independent public record of the incident.
State-sponsored AI operations
- A New York Times investigation found that actors linked to Iran and China, as well as private Israeli groups, used AI agents to automate influence campaigns (via KTUL). The systems created fake accounts, wrote posts and maintained fictional identities across Facebook, Instagram, X and TikTok.
- A Wall Street Journal opinion column examined an operation disclosed by Anthropic in which AI was used to monitor Tibetan communities and other dissidents.
- North Korea’s “Contagious Interview” campaign compromised more than 30,000 devices across 100 countries using fake job interviews (TechRadar). A joint report from agencies in Japan, the US, Germany and Australia attributed about $10 million in cryptocurrency theft to the campaign.
Britain’s new information-defence unit
- Prime Minister Andy Burnham announced a National Centre for Information Defence at the UN (BBC). Its remit is to “detect, attribute and disrupt” foreign disinformation. Burnham accused Russia of mounting an industrial-scale campaign against Britain.
- Nigel Farage compared the centre to Orwell’s “Ministry of Truth” (Express) and warned that it could be used against political dissent. The government says it will not police legitimate domestic debate.
- The Kremlin called the plan short-sighted and accused Britain of encouraging Russophobia (Yahoo/Reuters).
Deepfakes and synthetic media
- Fabricated videos of far-right protests spread on TikTok in Portugal, France, Germany, Ireland, Italy and the UK (Euronews). Some copied the presentation and branding of established news outlets.
- Fake videos alleged ballot fraud during German elections (Euronews). One claimed that the AfD had been removed from postal ballots. The material was linked to the pro-Russian Storm-1516 network.
- AI-generated images helped spread the absurd claim that Prince George has a secret older sister (Yahoo). The posts mixed genuine Royal Family photographs with invented people and events.
Election integrity
- Russia held its first parliamentary election since the 2022 invasion of Ukraine. People in Donetsk, Luhansk, Kherson and Zaporizhzhia voted in a State Duma election for the first time since Russia claimed to annex the regions (Bluewin). Voting also took place in Crimea.
- Ukraine and the EU rejected the votes conducted in occupied Ukrainian territory. The EU later sanctioned people involved in organising them (Bluewin).
- California enacted AB 1130, allowing fines of up to $5,000 per violation when influencers fail to disclose that a campaign paid for political content.
- Albania has no specific legislation or systematic monitoring for AI-generated election material (BIRN), despite its use during the 2025 parliamentary election. The country holds local elections in 2027.
Platforms and the press
- President Trump barred CNN, MS NOW and Politico from the White House and launched a 24-hour “Trump TV” YouTube channel (NPR).
- The ban lasted less than a week. A federal judge ordered the administration to restore the outlets’ access temporarily, saying it was likely unconstitutional.
- A New Mexico case over Facebook’s alleged misuse of personal data reached closing arguments (KVIA). The arguments returned to practices exposed by the Cambridge Analytica scandal, although this is a separate case.
Scams and fraud
- Attackers are placing malicious material online so that ChatGPT, Gemini and Google’s AI Overviews reproduce it (Dark Reading). The aim is to insert scams and phishing links into answers that users are inclined to trust.
- India’s Press Information Bureau debunked an altered video of foreign minister S. Jaishankar promoting an investment scheme (Latestly).
- Romanian former presidential candidate Călin Georgescu is under investigation over an alleged €1.1 million fraud (The Star). The case is separate from the controversy over Romania’s annulled 2024 presidential election.
Quick hits
- Pro-Kremlin accounts removed the context from a NATO anniversary video and presented it as “war propaganda” against Russia (France 24).
- Storm photographer Mike Olbinski says synthetic videos are using or imitating his footage (AZ Family), making invented storms and floods look genuine.
- Someone hacked a temporary traffic sign on Los Angeles’s 405 Freeway to display a pro-Iran website (Los Angeles Times).
Do you have Bad Aura?
A recruiter reading your feed may be looking for reasons to reject you. A scammer is looking for an opening. Bad Aura, from Aunoo, reads your public posts from both perspectives. It shows what a recruiter might regard as a liability and what a scammer could exploit.
About Bad Information
Bad Information is a weekly briefing on disinformation, coordinated influence operations and media literacy. It is researched and checked with Aunoo.
AuNoo AI evaluates news across five trust signals: the source, corroboration of its claims, whether the material appears human or synthetic, how it spread, and who owns the outlet. The purpose is not to award a single truth score. It is to show what can be established and what cannot. Use it with your AI Assistant!
Try it free → agentic.aunoo.ai
