Trellix, a global cybersecurity company that develops threat detection and response software, reported that threat actors are now selling malicious artificial intelligence tools in underground criminal forums. According to the company's research team, these tools lower the barrier for unskilled individuals to launch sophisticated cyberattacks.
During the first half of 2026, the security vendor identified several AI-related offerings spanning the attack lifecycle, from initial reconnaissance and exploit development to payload delivery, evasion and post-compromise operations. One identified tool, called APEX AI, requires attackers to input a targeted domain to generate a step-by-step plan for a ransomware attack. The software can also identify specific employees to target with spear-phishing campaigns.
The researchers noted a shift in how criminal communities interact with artificial intelligence.
“What distinguishes the current moment from earlier periods of AI hype in criminal communities is the shift from experimentation to commercialization,” Trellix stated. “The services documented here are not proof-of-concept (PoC) demonstrations or theoretical discussions. They are structured commercial offerings with pricing tiers, support channels, update cadences, and in some cases, customer reviews. This maturation mirrors the broader evolution of the cybercriminal economy, where specialization and service-based models have long been the norm for ransomware, initial access brokerage, and exploit development.”
The availability of these platforms alters the cybercriminal economy, paving the way for faster attacks from a broader range of threat actors.
“The integration of AI into this existing commercial infrastructure represents a qualitative change in capability, not merely a quantitative one,” the researchers wrote. “Tasks that previously required skilled human operators working across multiple tools and data sources are being compressed into single-prompt workflows. Evasion techniques that previously required manual tuning per target environment are being automated through per-build morphing.”
