While politicians tout AI’s potential for massive efficiency boosts, new threat intelligence data reveals a darker reality: criminal hackers have weaponized the same commercial tools to industrialize cybercrime. In recent months, AI-powered attacks evolved from experimental curiosity to full-scale operations targeting everything from zero-day vulnerabilities to supply chain infiltration.
The Speed of Escalation
The transformation from AI curiosity to cyber weapon happened faster than anyone expected.
Google Threat Intelligence Group tracked this transformation with unsettling precision. “There’s a misconception that the AI vulnerability race is imminent,” says John Hultquist, GTIG’s chief analyst. “The reality is that it’s already begun.” His team documented threat actors using AI to boost attack speed, scale, and sophistication across the entire intrusion lifecycle. One criminal group nearly deployed an AI-assisted zero-day exploit targeting two-factor authentication systems—vulnerabilities that bypass security measures requiring both passwords and secondary verification—in a mass exploitation campaign before coordinated disclosure prevented it.
Your Favorite AI Tools, Weaponized
The same LLMs powering workplace productivity are now driving criminal enterprises.
Criminal gangs and state-linked actors from China, North Korea, and Russia now routinely abuse commercial models—including Gemini, Claude, and OpenAI’s systems—to refine attacks. These aren’t basement hackers fumbling with code. They’ve built automated pipelines to create anonymized accounts, abuse free trials, and maintain persistent access to premium AI tiers while evading guardrails.
Beyond Script Kiddies
Modern AI-powered malware requires no human operator once deployed.
PROMPTSPY, an Android backdoor, uses Google’s Gemini API to navigate user interfaces and maintain persistence without human oversight. These “agentic workflows” execute multi-stage security tasks at machine speed, interpreting system states and making tactical decisions like seasoned operators. Supply chain attacks increasingly target AI environments themselves, creating footholds for ransomware and data theft—transforming AI infrastructure from productivity tool to attack vector.
The Defense Strikes Back
Security teams are fighting AI with AI in an escalating technological arms race.
Google’s own AI agents offer hope. Big Sleep systematically hunts unknown vulnerabilities while CodeMender generates automated patches using Gemini’s reasoning capabilities. Steven Murdoch, a security engineering professor at University College London, argues this evolution benefits defenders too, suggesting most future bug discovery will be “LLM-assisted” rather than purely adversarial.
Meanwhile, Anthropic chose not to release its Mythos model after internal tests revealed it could autonomously find zero-day vulnerabilities in major operating systems and browsers.
The emerging landscape resembles streaming’s disruption of entertainment—same underlying technology, completely transformed economics. Except here, both sides of the cybersecurity equation are racing to deploy AI faster than their opponents can adapt.
