Germany’s spy chiefs step into the light – and describe a shadow war
On 5 October the heads of Germany’s intelligence services took questions in a hearing open to journalists and to any citizen who had registered, and described Russian influence operations reaching down to state and local elections. Also in this issue: a quiet election in Latvia, how Russia primes AI chatbots ahead of the French election, and our study of the Czech disinformation scene. Welcome to issue 20 of Narrative Ops.
“Russia is waging a shadow war against us”: Germany’s spy chiefs testify in the open
Members of the Bundestag committee that oversees Germany’s intelligence services normally meet their chiefs in rooms protected against eavesdropping. Once a year since 2016 the session is public: journalists can attend, and so can any citizen who registers in advance. This year it took place on Monday, 5 October.
Martin Jäger, president of the BND, the foreign intelligence service, told the committee there is no sign of preparations for an immediate military attack. What Moscow is doing instead, he said, is sabotage, cyber attacks, espionage and disinformation: “Russia is waging a shadow war against us.” He described Germany as the obstacle Moscow has to remove to project power in Europe, because it supports Ukraine militarily, financially and economically, and warned that the country risks being drawn into a violent conflict.
Sinan Selen, who heads the BfV, the domestic service, described a change in targeting. Russian influence efforts used to concentrate on national elections; they now appear around state and municipal votes as well. Rather than sending its own officers, Russia recruits people already in the country for surveillance, transport or sabotage through digital channels, with social media central to the recruitment, he said. Selen called the wider effort coldly calculated destabilisation, and pointed to clear indications of Russian involvement in a drone with explosives found in August near a Ukrainian cargo plane at Leipzig-Halle airport.
Parliament is weighing a reform that would turn both services into what the government calls fully equipped intelligence agencies: wider powers to collect, store and automatically analyse data, and, for the first time, permission to disrupt sabotage and cyber attacks rather than only watch them. Jäger and Selen defended the plan. Konstantin von Notz, a Green MP on the oversight committee, questioned whether supervision will keep pace with the new powers, and expects the constitutional court to take up the reform.
Latvia’s election stayed quiet. The oddest finding came from the chatbots.
Latvia’s parliamentary election on 3 October was quiet. Prime Minister Andris Kulbergs’s centrist United List took 35.3% of the vote and 42 of the 100 seats, the largest win since independence, and is expected to form another coalition that supports Ukraine. The disinformation wave officials had prepared for did not come: Re:Baltica, an investigative outlet in Riga, found no increase in foreign disinformation before the vote, which matches the State Security Service’s own assessment, and claims of a stolen election, common after last year’s municipal vote, were rare. By election day the service had received 13 reports of possible violations.
The loudest fear campaigns were domestic, and the sharpest ran against immigrants. Re:Baltica found the National Alliance and several smaller right-wing parties portraying third-country migrants as criminals attacking Latvian women, including one fabricated sexual-assault story; one National Alliance candidate published an AI-generated video of a white man beating Indian immigrants. Third-country nationals other than Russians, Belarusians and Ukrainians make up about 1.5% of Latvia’s population and commit offences less often than local residents. Migration researchers told Re:Baltica this rhetoric can embolden real-world violence.
The campaign season still produced one finding worth keeping. A Latvian monitoring project reported in September by LSM, the public broadcaster, put the same questions about the country’s parties to ChatGPT, Gemini and Claude in Latvian and in Russian. The Russian-language answers favoured different parties than the Latvian ones and cited Sputnik, a sanctioned Russian state outlet, at least eight times as a source.
More voters now ask AI assistants to explain politics to them, and the researchers say such systems can be steered by flooding the web with content in a given language. Whether anyone did so before this election, they did not establish.
Before France votes in 2027, Russian operations are feeding the chatbots
France 24 counted 34 fake stories carrying its own brand over twelve months. A report published on 28 September by NewsGuard, compiled with France’s Viginum, Ukraine’s Center for Countering Disinformation and the research group Antibot4Navalny, traced 950 fake articles imitating 163 trusted news brands in 21 countries to three pro-Russian operations: Matryoshka, Storm-1516 and Doppelganger. In France, impersonations of known media grew 850% in a year, from 44 stories in the first half of 2025 to 421 in the first half of 2026, and eight of the 15 most imitated outlets worldwide are French, Libération, Charlie Hebdo, BFM, Le Figaro, Le Parisien and AFP among them. An 850% rise in one year looks like stockpiling: infrastructure built before the campaign that will need it.
The campaign is France’s 2027 presidential race. Three likely candidates who back Ukraine, Raphaël Glucksmann, Gabriel Attal and Édouard Philippe, have already been the subject of fabricated stories linked to pro-Russian groups. France is among the biggest contributors of aid to Ukraine, and NewsGuard’s Europe editor Chine Labbé calls its targeting no surprise.
Much of the output is written for machines. The Pravda network of some 280 pro-Russian sites published an estimated 6.3 million articles in 2025, volumes no human audience reads; the goal, Labbé says, is to whitewash the Russian narrative until it ends up “finding its way into chatbot responses”. One of its stories, dressed as Le Point, claimed 20,000 Ukrainian soldiers had been evacuated to France and deserted into criminal gangs; the official figure is just over 200 transfers, mostly wounded soldiers and sick children. When NewsGuard asked Google’s AI chatbot about it, the chatbot cited the fake articles as sources.
A fake that never goes viral still gets indexed. Shaping what an AI system answers is slow work: the material has to accumulate on the web, indexed and cross-cited, long before anyone asks the question, which is why the ground for 2027 is being prepared in 2026.
A study of 309,000 Czech posts shows influence campaigns running on everyday anger
Between December 2025 and May 2026, together with the Czech technology company Apify, we collected 309,527 posts, videos, articles and comments from Czech social media and news sites and sorted them into 4,462 topics.
Open praise of Russia barely registers anywhere in the dataset. What spreads starts from a grievance that is real – energy bills, inflation, a corruption story – and swaps the culprit: expensive energy becomes “Brussels is to blame”, inflation becomes “the West is robbing us”. We call the loss-of-sovereignty narrative a gateway: it links migration, the Green Deal, energy prices and support for Ukraine into one message, that decisions about Czech life are no longer made in Prague.
The content that resonates is built on grievances that are, in the words of Liudvikas Andriulis, Repsense co-founder, “real and verifiable, which makes them far harder to counter than fabricated lies”. To see who organises the spread, you have to look at coordination, not content. We found it in a network campaigning for Czech withdrawal from the EU, which moved identical posts across the same circle of Facebook groups.
Meloni applies to trademark her own voice against AI clones
Italian prime minister Giorgia Meloni has applied to register her voice as an EU trademark, Reuters reported on 5 October. If the application succeeds, anyone using a cloned version of it without permission could be ordered to take the material down and pay damages. The filing follows a run of AI videos and audio that imitate European politicians.
Paid ads told Brazilians not to trust the voting machines days before the vote
In the last weeks before Brazil voted on 4 October, ads on Facebook and Instagram, some paid for by senate candidates, told voters the electronic voting machines could not be trusted; the research institute Democracia em Xeque counted the daily average of such posts doubling from August to September. After BBC News Brasil asked about the ads, Meta removed several and temporarily suspended all political advertising aimed at Brazil. Flávio Bolsonaro and Luiz Inácio Lula da Silva meet in a run-off on 25 October.
Fake NATO deaths, Epstein claims and drones: how Russia targeted the Baltic elections
A video styled as a Euronews report claimed a Ukrainian soldier had shot three NATO servicemen during the Baltic Trust 26 exercise in Latvia, then himself. It circulated for weeks before Latvia’s 3 October vote through dozens of accounts with a combined audience above 1.85 million. Latvia’s armed forces called it a coordinated disinformation campaign and Euronews confirmed it never produced the footage; researchers counted the same video rebuilt in eight versions, including Arabic and Spanish edits. Other fakes in the same wave tied Estonia’s outgoing president and Lithuania’s Gitanas Nausėda to Jeffrey Epstein through a cloned EUobserver site, and invented a Ukrainian drone strike on Portuguese F-16s at Ämari air base.
Romanian outlets cited chatbots as proof a sea drone was Ukrainian
After a naval drone was destroyed near Romania’s Neptun Deep gas platform in August, Romanian sites, Facebook pages and Telegram channels declared it “undoubtedly” Ukrainian because ChatGPT, Gemini and Grok said so; one article drew more than 3,000 reactions. Context.ro reran the test with the same official photos and a neutral prompt: ChatGPT pointed to a vessel from the British-derived MAST-13 family, Grok to a previously unknown type most likely of Russian origin, and Gemini to a Ukrainian MAGURA V5.
Last week we reported the one-page memo in which US defence secretary Pete Hegseth ordered Cyber Command and the NSA to work on foreign interference in the midterms.
Update: four weeks before the vote, state election officials say they are not hearing from Washington. A US intelligence assessment shared with Congress names Russia as the most active adversary, using online disinformation to sow distrust in the results, with China targeting individual races; it found no sign that any foreign actor can interfere with voting machines or vote counts, the Washington Post reports. Republican officials in Utah and Idaho said briefings and systems testing had stopped and that CISA, the civilian cyber agency, has had only acting directors. CISA told the paper it is implementing its new election infrastructure security plan and offers states threat information sharing, technical expertise and vulnerability scanning.
Get Narrative Ops in your inbox
Powered by Repsense narrative intelligence
Powered by Repsense narrative intelligence