AI Chatbots Repeated Russian Propaganda Claims in 70% of Test Answers, Study Finds, as Hackers Hijack War Alert Apps

Economic Warfare & Trade Policy

People shelter in the Kyiv Metro during an air alert on Jan. 20, 2026 (file photo). Photo: Kyiv City State Administration, Wikimedia Commons, CC BY 4.0.

Two leading AI chatbots repeated disputed or false claims from Russian sources in 70% of their answers to questions about those claims, according to a study by the Global Disinformation Index and analytics company Sy.nexus, Ukrainska Pravda reported. The researchers tested Anthropic’s Claude Sonnet 5 and Mistral AI’s Mistral Large on “data voids,” events for which little independent, verified information exists online while state-controlled outlets publish heavily.

Key figures: 70% AI answers that repeated disputed Russian claims · 140 of 280 claude answers whose searches surfaced tass · 11 tankers iran’s irgc navy says caught fire since oct. 25

One test claim came from Russia’s Defense Ministry, which said its forces had captured the village of Vilne in Ukraine’s Dnipropetrovsk region. Both models reproduced the Russian version. Mistral sometimes stated the village had been captured, while Claude more often added that the information came mainly from Russian sources and that no confirmation could be found from Ukrainian authorities, the Institute for the Study of War or the Ukrainian monitoring project DeepState. A similar pattern appeared with a claim that a Ukrainian drone had attacked a civilian bus in Russia’s Belgorod region.

The difference lay in presentation. Claude questioned or challenged the claims in 60% of its responses, repeated them without assessing credibility in 12.2% and presented them as fact in 27.8%. Mistral issued caveats in just 19.8% of cases, and almost half of its answers presented the claims as facts. Claude ran an average of 3.6 web searches before answering, compared with 1.2 for Mistral, and so encountered Russian state outlets more often. The study’s authors linked the problem to retrieval-augmented generation, in which models search the web for recent events, and to “generative engine optimization,” in which websites tailor content to appear in AI search results.

False alarms on both sides of the front

Russia and Ukraine both reported cyber incidents affecting their emergency alert apps early Monday, with false warnings sent to users in both countries, Anadolu Agency reported. Ukraine’s Air Force said its Air Alert app had likely been subjected to “hostile interference,” adding on Telegram: “There is no radiation threat.” Ukraine’s Center for Strategic Communications blamed Russia, saying users had received “strange messages,” including a Russian-language message referring to the Oreshnik missile. Hours later, Russia’s Emergencies Ministry said its mobile app had been hit by a “massive cyberattack” that allowed attackers to send false information, TASS reported !Unverified claim, TASS (Russian state media). The ministry said Russia’s broader warning system of sirens, text messages and broadcasts was unaffected, according to Anadolu. Neither side publicly identified who was behind the Russian breach. Today’s frontline developments are in our Russia-Ukraine war report.

Competing claims in the Strait of Hormuz

Iran’s Islamic Revolutionary Guard Corps Navy said 11 “violator tankers” had caught fire in the Strait of Hormuz since Oct. 25 and that some had sunk, TASS reported !Unverified claim, TASS (Russian state media). The claim has not been independently confirmed. The U.K. Maritime Trade Operations center has reported individual incidents without identifying vessels; on Nov. 1 it relayed a Mayday call reporting that a ship had been hit, without giving details of casualties, damage or the source of the projectile, Anadolu reported. Today’s verified tanker incidents are in our Middle East report.

Russia’s Federal Security Service said it had seized a batch of FPV drone goggles intended as humanitarian aid for Russian troops that it said Ukrainian special services had fitted with improvised explosive devices, TASS reported !Unverified claim, TASS (Russian state media). The claim could not be independently verified. A Russian military commentator cited by TASS said Ukraine lost 35,840 troops killed or wounded in October, the agency reported, while Ukraine’s General Staff said Russian forces suffered a record 2,500 casualties on Nov. 1, or one every 35 seconds, The Kyiv Independent reported. Neither figure can be independently verified.

Hacks, zero-days and a fake crash photo

Ivanti released patches for two Connect Secure remote-access flaws, CVE-2026-111357 and CVE-2026-118527, that attackers had chained in zero-day attacks since at least Oct. 1, BleepingComputer reported. The flaws were exploited alongside an authentication bypass Ivanti patched last month. Network defenders can check the U.S. CISA catalog of known exploited vulnerabilities. Earlier coverage of state-linked cyber threats is on our security threats page.

Binance co-founder Yi He denied an online rumor of a romance between one of the crypto exchange’s engineers and an alleged North Korean agent, after the exchange reported a cyberattack on Oct. 29. “This story is dramatic, but it is not true,” she wrote on X, TASS reported !Unverified claim, TASS (Russian state media). The hack is covered in our Binance report.

In the United States, a widely shared image of the Learjet that crashed in Palm Beach, Florida, on Nov. 1 showing a missing wing was edited with AI tools, Lead Stories found. Google’s SynthID detection tool found an invisible watermark consistent with AI modification, and video of the crash shows both wings.

By the numbers

The GDI researchers also tracked which outlets appeared in the search results Claude used to write its 280 answers. Mistral’s searches did not surface these outlets, according to the study.

Watch for forensic findings on the alert-app breaches and independent confirmation, or refutation, of the IRGC’s tanker claims.

Sources

• Ukrainska Pravda: Can AI tell the truth? How Russian propagandists are influencing ChatGPT
• Anadolu Agency: Russia, Ukraine report cyberattacks on emergency alert apps
• TASS: Russian Emergencies Ministry's mobile app hit by massive cyberattack
• TASS: Iran's IRGC Navy says 11 tankers violating Hormuz passage rules catch fire since Oct 25
• Anadolu Agency: Vessel reportedly struck by unknown projectile in Strait of Hormuz: UK maritime agency
• TASS: What we know about seizure of booby-trapped FPV goggles intended for Russian troops by FSB
• TASS: Kiev loses nearly 36,000 troops, mercenaries in October — military expert
• The Kyiv Independent: Russia lost 1 soldier every 35 seconds on Nov. 1, Ukraine's General Staff says
• BleepingComputer: Ivanti patches two more Connect Secure zero-days chained in attacks
• Cybersecurity and Infrastructure Security Agency: Known Exploited Vulnerabilities Catalog (JSON feed)
• TASS: Binance co-founder refutes version of engineer's romance with alleged North Korean agent
• Lead Stories: Fact Check: AI-Enhanced Image NOT Actual Photo Of Learjet Crash In Florida

Additional sources

• Mehr News Agency: Two non-compliant oil tankers detonate, burn in Hormuz Strait
• The Kyiv Independent: Russia exploits ceasefire to mask military movements