Moldova's next election is already under attack, AI fakes chase the World Cup's final days, and two operations collide on the same Filipino pages
This week's issue runs from Moldova's next elections to the World Cup's final matches, with stops in Manila, Paris, Canberra and a Munich courtroom. Welcome to issue eight of Narrative Ops.
Moldova's intelligence chief says the interference machine has moved to TikTok
Russia is already working on Moldova's next elections. Alexandru Musteață, who runs the country's Information and Security Service, told Ziarul de Gardă this week, in an interview relayed by Deschide.md, that his service has logged a sharp rise in disinformation and propaganda activity this year, run with different methods than the ones used against the 2024 presidential vote and last September's parliamentary election. Telegram used to be the main pipe. It is TikTok now.
They have very large teams of specialists, well paid, who work against us every day.
– Alexandru Musteață, director, Moldova's Information and Security ServiceThe targets Musteață listed are the army, the security sector, the public-order bodies and the justice system. He was explicit about the point of hitting them: drive down trust far enough that by the next electoral cycle the ordinary citizen doubts whether any state institution works. That is a slower objective than swinging a ballot, and a more durable one. He also described penetration of the political class itself, through recruitment and promotion of individual politicians and through illegal financing of people acting in Moscow's interests.
Moldova is better prepared than it was a year ago, he said, and then immediately warned against relaxing. Two election cycles have made the country Moscow's testing ground for this work, and Chișinău has spent the past month closing doors: the agreement letting Russia's "Russian House" operate in the capital was allowed to lapse, and the Ministry of Culture published a list of Russian artists whose concerts it advises against on national-security grounds.
The wave Musteață describes is one we measured. Around Moldova's September 2025 elections, Repsense analysed some 250,000 media items – web, TV, radio and seven social platforms, tracking 28 parties and 104 politicians. Pro-Kremlin narratives dominated: war fears, anti-EU messaging, Moldova as a pawn, and fake stories claiming Romania wanted to annex the country or that NATO troops were already stationed there – amplified by AI-generated personas, duplicate accounts and synchronised posting. And of 13,673 highly relevant short videos, TikTok and Instagram drew the most engagement: the machine he says has moved to TikTok was already busiest there.
Read the Moldova case study →The World Cup has turned into a proving ground for synthetic media
Five days of the tournament remain, and the scam-ad wave we covered in last week's issue has not slowed. In Revelum's count, a single campaign has run 254 separate ads built on Ronaldo's face and voice, and once the football started on June 11, new scam ads began appearing faster: the daily rate of ads impersonating Luis Díaz rose 413 percent.
Not all of it is a sales pitch. The European Broadcasting Union and Euronews have been cataloguing fabricated match footage, invented player quotes, AI-managed fan accounts and doctored photographs – among them an image of a supporter resembling Hitler at the Germany–Curaçao game that ran on Facebook, Instagram, Threads and Reddit in English, Spanish and Russian. One post carried it past three million views. Another manipulated image put the UK's outgoing prime minister Keir Starmer in a Croatia jersey next to former deputy prime minister Angela Rayner; a reverse image search traced it back to a real photo Rayner posted on X in June 2024.
The scam ads will be gone by August. The habit they leave behind is the part worth worrying about: once an audience learns that footage this convincing can be manufactured, real footage of real misconduct becomes dismissible – the liar's dividend researchers have been warning about for years, now playing out at the scale of a 48-team tournament across 16 host cities. FIFA and several national federations have fallen back on digital forensics and content credentials to authenticate their own announcements.
Two operations hit the same Filipino pages from opposite directions
Two influence operations, one Filipino and one Chinese, spent this spring working the same handful of Facebook pages without any apparent relationship to each other. A DFRLab investigation published June 30 pulls them apart. The domestic operation it links with moderate confidence to the 2nd Civil-Military Operations Battalion of the Philippine armed forces, on circumstantial evidence: military accounts in the engagement patterns, profiles wearing Civil-Military Operations insignia, six interlinked pages built between February and April.
The Filipino operation wanted the activists discredited. After the military killed two university students in an April clash with NPA rebels, it red-tagged the League of Filipino Students, the labour federation Kilusang Mayo Uno and their organisers as communist front groups. Spamouflage wanted the audience. Fifty inauthentic profiles, part of a network Meta attributes to Chinese law enforcement, filled the same comment threads with AI-generated images and formulaic posts about corruption and Marcos's health, talking over a local conversation about two dead students in order to attack the president and elevate his rival, Vice President Sara Duterte. Some of those accounts called for more Molotov cocktails at the protests.
Meta took down Philippine military assets in 2020 and the same enforcement gap is still open six years later. Both operations marshalled hundreds to thousands of profiles to coordinate reactions, which the DFRLab says should be trivial to flag at the platform level. Activist pages that restrict comments to long-term followers were beaten by profiles that simply liked the page first. An activist told Amnesty International that reporting red-tagging pages does not stop the harassment.
France wants to triple the price of election lies
French prime minister Sébastien Lecornu told the Senate on July 8 that he will bring a bill to the Council of Ministers in late July tripling the penalties for producing false content during electoral periods, a window he called "sacred" for democracy. The announcement answered a question from senator Claude Malhuret about AI in the coming presidential campaign – doctored videos, cloned voices, invented characters and bot-driven viral lies. Current penalties, Lecornu argued, are not sufficiently deterrent.
The bill reaches beyond sentencing. An emergency judicial procedure that can order false content taken down – today limited to presidential, legislative and senatorial votes – would be extended to all local elections, and a permanent "public information commission" would be created by decree, described as an alert body for the press, judges and citizens when interference is detected, not a censor. The text builds on France's 2018 law against the manipulation of information and sits alongside the EU's Digital Services Act.
The objection is the oldest one in the fake-news file: who decides what is false. Free-speech advocates warn that vague standards and state-appointed bodies risk chilling legitimate speech in the charged weeks of a campaign, and France's own record cuts both ways – an earlier rule ordering extremist content removed within an hour was later curbed by the constitutional court.
RT's new X account clears 6 million views under sanctions
NewsGuard documented RT's @RT_on_X account gathering roughly 6 million views within five days of its late-June debut. The European Commission says its sanctions cover every distribution channel, including apps and platforms.
A Russian operation is piping sanctioned media into the fediverse
CheckFirst identified "Roska Bridge," a network of accounts that republishes content from EU-sanctioned Russian outlets on Mastodon and Bluesky, where enforcement barely reaches. It abuses Brid.gy, a legitimate tool for syncing posts between decentralised platforms, publishes in monthly bursts, and shares structural fingerprints with the Pravda network.
Vietnam-run pages faked Australian news brands for ad money
The Disinformation Observer detailed Facebook pages impersonating The Australian and others to fabricate political stories, one falsely claiming senator Pauline Hanson collapsed in parliament. Facebook's transparency data showed one page run by twelve people in Vietnam, with another also managed from Indonesia.
X will DM you when a post you boosted gets corrected
Elon Musk announced that Community Notes will message users whose liked, replied-to or reposted posts later receive a correction, with no launch date yet. The change targets the system's known weakness: studies put the share of proposed notes that never get published at 85–90 percent, and the ones that do land often arrive after the damage is done.
We covered Reddit's spam numbers and the rise of "generative engine optimization" – content seeded to be quoted by ChatGPT and Gemini, written for the machine that answers on the reader's behalf.
Update: If you write for the answer engine, someone owns the answer. The Regional Court of Munich has now said who. Its May 28 injunction (case 26 O 869/26) found Google directly liable for AI Overviews that tied two Munich publishers to scams and subscription traps, drawing connections that appeared in none of the cited sources. An overview producing "independent, new, and substantive statements" gets no liability shield, the court held, and Google's defence that users can click through and check was rejected outright. The order is a temporary injunction, not a final ruling; Google says it is reviewing the decision and is expected to appeal.
Get Narrative Ops in your inbox
Powered by Repsense narrative intelligence