AI-generated malware poses little real-world threat, contrary to hype

On 5 November 2025, Google revealed five recent malware samples built with generative AI, but their outcomes were far below professional malware development and pose little real-world threat. The samples—PromptLock, FruitShell, PromptFlux, PromptSteal, and QuietVault—were easy to detect even by basic endpoint protections and relied on methods already seen in prior malware, according to Google’s assessment.

PromptLock, which was also described in an academic study as showing the feasibility of AI for ransomware, was noted to have clear limitations: it omits persistence, lateral movement, and advanced evasion tactics, serving mainly as a demonstration. The report references ESET’s discovery of PromptLock and its portrayal of it as the first AI-powered ransomware.

It also notes that while industry voices such as Anthropic, ConnectWise, and OpenAI have discussed AI-assisted cybercrime, Google found no evidence of automated breakthroughs or credible near-term enhancements to threat capabilities. Ultimately, the takeaway is that AI-generated malware remains largely experimental, with the biggest threat continuing to rely on traditional methods.