Malware: Public administration, Ukraine 2025
In November 2025, the Google Threat Intelligence Group described for the first time malware that queries a language model while in use.
- Incident · 11.2025 · Ukraine (PROMPTSTEAL); PROMPTFLUX only in testing, Ukraine
- Evidence: confirmed by an authority
- Malware
- AI link
As of
What happened
PROMPTSTEAL of the Russian group APT28 used this to obtain Windows commands, collected system and document data and sent it to the attackers. Google blocked the Gemini access associated with PROMPTFLUX.
Role of AI: PROMPTSTEAL (APT28) has Qwen2.5-Coder generate commands for data collection via Hugging Face; PROMPTFLUX has its code rewritten by Gemini, but according to Google is still experimental.
- Affected
- Company (anonymous)
- Industry
- Public administration
- Loss
- no reliable figure
- Business interruption
- unknown
- Ransom demanded / paid
- no / no
- Data affected
- true
- Bankruptcy
- no
- Entry point
- PROMPTSTEAL disguises itself as an image generation program
What would cyber insurance have paid?
That depends on the terms. Whether those affected were insured: unknown.
-
Costs of getting data, software and systems running again
Mostly covered 6 of 9 business products cover it fully, as an add-on or with limits.
-
Immediate help in an emergency
Hotline, IT forensics, crisis PR and legal advice
Mostly covered 7 of 9 business products cover it fully, as an add-on or with limits.
Overview based on public policy terms, as of 06.11.2026. Only your policy and its general terms and conditions (GTC) are binding. No confirmation of cover.
What we can learn
Malware that continuously renews its code using AI requires behaviour-based detection instead of signatures alone.
Sources
- GTIG AI Threat Tracker: Advances in Threat Actor Usage of AI Tools (opens in a new window) Google Threat Intelligence Group · 2025-11-05
«PROMPTSTEAL novelly uses LLMs to generate commands for the malware to execute rather than hard coding the commands directly in the malware itself.»
- Google warns of new AI-powered malware families deployed in the wild (opens in a new window) BleepingComputer · 2025-11