Cybercriminals are using artificial intelligence to make attacks faster and deception harder to spot. AI agents can carry out parts of an intrusion with limited human direction, while deepfake voices and video can make a fraudulent request appear to come from a trusted executive. Recent cases show different uses of AI rather than one universal attack method.
In a campaign disclosed by Anthropic in November 2025, a group it identified as GTG-1002 used Claude Code to support cyber espionage against roughly 30 organizations.
Anthropic assessed the group with high confidence as Chinese state-sponsored. Its report says the AI performed about 80% to 90% of the tactical work, while human operators made a small number of key decisions.
Deepfake AI Attacks Scale Deception
The AI-assisted work included reconnaissance, vulnerability discovery, exploitation, credential harvesting and analysis of stolen data.
That does not mean the attack ran without people, operators selected targets and approved important steps. It does show how automation can let attackers test and pursue several targets with far less hands-on effort.
AI is also appearing inside malware. In February 2026, ESET described PromptSpy, Android malware that uses Google’s Gemini to interpret screen elements and guide actions that keep the malicious app in the device’s recent-app list.
PromptSpy also abuses Android’s Accessibility Service to obstruct removal with invisible overlays. Its main purpose is to deploy a module that gives operators remote access to the device.
ESET said the campaign appeared financially motivated and primarily targeted users in Argentina; it did not attribute the operation to a Chinese state actor.
These examples matter because fixed detection rules may struggle when an attacker changes tactics or when malware adapts to what appears on a screen.
Defenders still need to watch for suspicious access, unusual account behavior and unauthorized use of powerful device permissions.
Automation is only part of the threat. Deepfakes can exploit the moment an employee decides whether a request is genuine.
In a case reported in 2024, an employee at engineering firm Arup’s Hong Kong office joined a video call that appeared to include senior colleagues.
The employee then made fraudulent transfers totaling about $25 million. The incident was reported in 2024, not early 2026, trellix said.
The lesson is not that every video call is fake. It is that a familiar face or voice should not, by itself, authorize a high-value payment.
A separate call to a previously verified number and approval from another authorized employee can help break the chain of impersonation.
Organizations should apply those checks even when a request appears urgent or comes through a trusted account.
Join 16,000+ SOC teams using ANY.RUN to streamline threat investigations and reduce manual effort. Explore for your team
