🚨AI-Powered Cyberattacks Surge Among Hackers
AI is now a standard tool for cybercriminals
TL;DR
AI is increasingly being used by hackers, from sophisticated groups to less tech-savvy actors. Four new threat groups used AI models like Claude and Deepseek to autonomously carry out attacks over months.
AI adoption among cybercriminals is on the rise, with even unsophisticated hackers incorporating agentic AI into their campaigns. Between May and August, four previously unreported threat groups leveraged AI to autonomously carry out attacks, highlighting the shift towards AI as a standard tool in hacking. These groups used both open- and closed-weight models, often opting for older versions to avoid detection. The more sophisticated actors built complex AI infrastructures, including 'Supervisor' and 'Executor' agents, while less skilled groups relied on basic, publicly available skills. This trend underscores the evolving nature of cyber threats, where AI is not just about exploiting new vulnerabilities but enhancing the speed and scale of existing attacks.
Key Points
Four new threat groups used AI models like Claude and Deepseek to autonomously carry out attacks between May and August.
Two groups exhibited sophisticated AI 'harnesses', while the other two showed less technical prowess, using basic publicly available skills.
Chinese-speaking actors used Claude Opus 4.8 and other models, setting up fallbacks to other AI tools if Claude was unavailable.
Spanish-speaking actors targeted Argentine healthcare entities, compromising at least 15 organizations and stealing medical records.
Russian and Indonesian-speaking actors compromised multiple government domains and private-sector targets using basic publicly available skills.
Why It Matters
If you're in cybersecurity, the rise of AI-powered attacks means you need to rethink your threat detection strategies. Hackers are using AI to autonomously carry out attacks, making traditional detection methods less effective. For instance, the Chinese-speaking actor used Claude Opus 4.8 and set up fallbacks to other AI tools, highlighting the need for more robust and adaptable security measures.
Comments
Be the first to comment
Enjoyed this article?
Get it daily. 7am. Free. Reads in 5 minutes.
Join 3,569 builders reading daily.