Kremlin deepfakes put surrender pleas in Ukrainian lawmakers' mouths
Two AI-generated videos are doing the rounds on pro-Kremlin Telegram channels, showing Ukrainian figures apparently conceding the war — and they're a case study in how little it takes to make propaganda stick.
Pro-Kremlin Telegram channels are spreading deepfake videos of two Ukrainian figures calling for peace talks, according to the disinformation tracker NewsGuard. The clips feature Andriy Levus, a former deputy head of Ukraine's Security Service, and Ivan Yunakov, an MP from Zelenskyy's party. Levus appears to plead for negotiations; Yunakov mourns lost territory. Both videos carry the usual tells — unnatural blinking, flat voices, stiff expressions — and together they racked up roughly 130,000 views in two weeks before being flagged.
Why it matters: debunking is not a win condition here. If nobody catches a fake, its message spreads unchallenged; if fact-checkers do catch it, the episode still erodes trust in every clip of a real official. The lie wins either way, which is exactly what makes this genre durable.
It's also a reminder of how far the tools have traveled since 2022, when a crudely spliced video made President Zelenskyy appear to order surrender — a fake so clumsy it became a punchline within hours. Today's version needs no studio, just a voice model and a face swap, and NewsGuard notes many similar clips likely go undetected. The countermeasures haven't kept pace: platform takedowns move in days while generation takes minutes, and provenance standards like C2PA only cover content that carries them — we reported last week that even those watermarks can be forged on rooted hardware.
The takeaway isn't that viewers should distrust everything; it's that verification now has to happen at the claim level, not the pixel level. A clip looking real proves nothing either way.
What to watch: whether Ukrainian authorities formally attribute the videos, and whether the next wave moves from Telegram to platforms with algorithmic reach.
Seen a suspicious "official" clip circulating? Tell us in the comments.
SenseTime has posted its first-ever profit as a listed company — RMB 620 million (about $87 million) for the first half of 2026, per its interim results. The Chinese AI champion, unprofitable since its 2021 Hong Kong IPO, credited its pivot to generative AI infrastructure: daily token serving on its SenseNova compute platform hit 2.4 trillion in July, up roughly 22x year over year, with recurring revenue and gross margin both improving.
One honest caveat: the company attributes part of the turnaround to fair-value gains on its ecosystem investments, so the milestone is partly accounting as well as operations. Still, token volume growth is real demand, not bookkeeping — and it echoes DeepSeek's leaked financials this week, which showed frontier-scale AI economics running far leaner than Silicon Valley assumes. The "token factory" model is quietly becoming China's answer to the GPU capex arms race.
QueryStory emerged from stealth with a $6 million seed round to fix enterprise AI's trust problem, raising at a $60 million valuation from Brightmind Ventures and New York Life Ventures. Founded by ex-Google Chronicle security veterans, the platform runs analytics through LLM agents but attaches a confidence indicator to every analysis and routes flagged results to human review — selling verifiable answers rather than raw intelligence. It's an early entry in what looks like a coming wave: trust-layer tooling for enterprises tired of gambling on chatbot output.
Sources: The Decoder · NewsGuard · Leiphone · Bamboo Works · TipRanks · TechCrunch
