Media & Disinformation Update: AI & Manipulation
Deepfakes, intellectual property, and AI slop
AI has become both a tool for manipulation and a challenge for those trying to verify what is real. From image detection systems that incorrectly label authentic images as fake, to leading chatbots repeating state-backed propaganda, reliability of information is under pressure from both humans and machines.
This edition of the Media & Disinformation Update from perspicacity features:
The AI detection tools falsely labelling genuine content as synthetic
A major European AI chatbot’s susceptibility to Iranian war propaganda
Claude’s increasing vulnerability to state-backed disinformation
How major technology platforms are promoting AI-powered ‘nudify’ apps
Wikipedia’s ban on AI-generated articles
A landmark challenge in Brazil to the use of journalism content by AI platforms
Europe’s efforts to regulate harmful AI applications while delaying wider AI safeguards
perspicacity is committed to sharing investigative, fact-based reporting that meets the highest standards of journalistic integrity. To access the full stories summarised below, click on the links.
AI AUTHENTICATION
Leading AI image detection tools mislead online users, often declaring authentic content fake
NewsGuard, 8 May 2026
AI detection tools are known to sometimes misidentify AI-generated images as real, but an audit of leading AI detection tools has found another concerning weakness: three of the five leading tools tested thought real images were fake.
According to NewsGuard’s findings, the tools collectively declared authentic images to be AI-generated 13% of the time. This vulnerability has potentially significant consequences, empowering bad actors to dispute reality by citing a detection tool to argue that a given image is fake - and unfairly maligning those posting the real images.
ScamAI, which claims “Industry-leading accuracy” on its website, declared that six of 15 authentic images were AI-generated - a 40% fail rate. Hive and Sightengine correctly identified all 15 authentic images as real.
AI SLOP
Anthropic’s AI chatbot is learning more on Russian and Iranian propaganda sources, NewsGuard audit finds
NewsGuard, 4 May 2026
Anthropic’s AI chatbot Claude repeated false claims 15% of the time when it was asked about pro-Kremlin falsehoods - and cited Russian state-affiliated media every time - a NewsGuard audit found.
This marks a sharp increase compared to seven previous audits between March 2025 and February 2026, when Claude cited Russian state-affiliated sources only 4% of the time on similar prompts.
The increase suggests that Claude has become more vulnerable to state disinformation campaigns.
NewsGuard has been auditing Claude and nine other chatbots since July 2024, and Claude has consistently been among the top performers.
Related:
Why the racist AI slop industry is booming
The Bureau of Investigative Journalism, 19 May 2026
The devout Muslim making a living from Islamophobic AI slop
The Bureau of Investigative Journalism, 5 May 2026
FAKE NEWS
Mistral’s Le Chat, Europe’s leading AI chatbot, repeats falsehoods half the time when prompted on state-sponsored Iran War disinformation
NewsGuard, 28 April 2026
Europe’s leading home-grown chatbot is also a leading propaganda spreader. When prompted on false claims about the war in Iran advanced by Russian, Chinese, and Iranian state-aligned networks or outlets, Paris-based Mistral AI’s chatbot, Le Chat, repeated falsehoods 50% of the time in English and 57% of the time in French, an April 2026 NewsGuard audit found.
The findings suggest that Le Chat, the most prominent European-made chatbot, is vulnerable to state-sponsored disinformation, potentially increasing its spread.
2026 Iran War: 103 false claims and counting
NewsGuard, 8 May 2026
NewsGuard has identified 103 false claims related to the Iran conflict that collectively have attracted hundreds of millions of views. Three trends are apparent: the false claims push a pro-Iran agenda; they have evolved from using misrepresented ‘real’ photos and videos to using AI-generated imagery; and increasingly they involve assertions that accurate stories published by authentic media are AI-generated disinformation.
Wikipedia bans AI-generated articles
The Verge, 26 March 2026
Wikipedia will no longer allow editors to write or rewrite articles using AI. The update, which was added to Wikipedia’s guidelines late last week, cites as the reason for the ban the tendency for AI-written articles to violate “several of Wikipedia’s core content policies.”
INFLUENCE OPERATIONS
TikTok accounts use AI-generated videos to incite anti-government protests in Moldova
Digital Forensic Research Lab, 3 April 2026
A network of at least 25 TikTok accounts appeared to co-ordinate the dissemination and amplification of AI-generated content targeting Moldova’s ruling Party of Action and Solidarity and the country’s president, Maia Sandu. It also sought to mobilise residents to participate in protests scheduled for 25 March, which ultimately did not take place.
The seeder accounts primarily published AI-generated videos that depicted fabricated protest scenes in Moldova. The videos often featured synthetic crowds and called for citizens to join anti-government demonstrations.
INTELLECTUAL PROPERTY
Unprecedented proceeding against Google opens way to fair compensation for journalism in Brazil
Reporters Without Borders, 4 May 2026
Brazil’s competition authority has for the first time opened proceedings against a tech giant to investigate the harm it has caused to journalism. The proceedings were launched into Google News and Google’s search engine, whose features use generative artificial intelligence.
The members of the Administrative Council for Economic Defense, Brazil’s anti-trust authority, unanimously concluded that there is evidence of anti-competitive conduct by Google in its use of journalistic content, including services that use generative AI.
From displaying short excerpts from news articles in search results to generating AI-powered responses using content from the press, Google has increasingly extracted value from journalistic content, reduced traffic to news sites and undermined their ability to sustain themselves through advertising revenue.
NON-CONSENSUAL INTIMATE IMAGES
Apple and Google steer users to nudify apps
Tech Transparency Project, 15 April 2026
Apple and Google are helping users to find apps that create deepfake nude images of women, a Tech Transparency Project investigation has found, showing how the platforms are key participants in the spread of AI tools that can turn real people into sexualised images.
The Tech Transparency Project revealed in January that the Apple and Google app stores each hosted dozens of ‘nudify’ and undressing apps that can digitally ‘strip off’ someone’s clothes. This latest investigation found that the app stores’ search and advertising systems actually point users to such apps, giving them increased visibility.
EU backs nude app ban and delays to landmark AI rules
The Verge, 26 March 2026
European lawmakers have voted to delay key parts of the EU AI Act, the bloc’s flagship law for regulating artificial intelligence, while also backing proposals to ban nudify apps. The proposals would push back looming deadlines for watermarking AI-generated content and high-risk AI systems.
The measures, approved by a large majority in the European Parliament, would push back compliance deadlines for developers of high-risk AI systems — those deemed to pose a “serious risk” to health, safety, or fundamental rights — until December 2027.
Companies developing AI systems covered by specific safety rules in sectors like toys or medical devices would have even longer to comply, with a proposed deadline of August 2028. Rules requiring providers to watermark AI-generated content would also be delayed until November 2026. All of these measures had originally been set to take effect in August 2026.
Members also backed proposals to include a ban on nudify apps in the revised AI Act, following widespread outrage in the EU over the flood of Grok’s sexualised deepfakes on X earlier this year.