Cybercriminals have found a disturbingly effective new attack vector, and it targets the very AI tools companies rushed to deploy. CrowdStrike’s 2026 Global Threat Report reveals that prompt injection attacks hit more than 90 organizations, with adversaries exploiting large language models to steal credentials and, in at least one case, cryptocurrency assets.
The report, published on February 24, 2026, paints a picture of an AI arms race where defenders are losing ground fast. AI-enabled attacks jumped 89% year-over-year, and the average time for an attacker to move laterally through a compromised network, what CrowdStrike calls “eCrime breakout time,” dropped to just 29 minutes in 2025. The fastest observed incident? 27 seconds.
How prompt injection became the top AI threat
The attack technique earned the top spot on the OWASP Top 10 for LLM Applications in 2025, ranking as the number one risk factor. CrowdStrike says it now tracks over 180 techniques related to prompt injection and AI exploitation. In response, the company launched Falcon AIDR in December 2025, a tool specifically designed to detect and mitigate prompt injection attacks and unsafe AI outputs.
The crypto connection: $175K drained via Morse code
