Cybercriminals gave AI a go — and came away disappointed, study finds

A University of Edinburgh study of more than 100 million cybercrime forum posts found little evidence that AI has materially improved hackers’ attacks or tools.

Most reported AI use was limited to basic automation, coding help, phishing, SEO fraud, and social media bot tasks.

Researchers also saw attempts to bypass safeguards in mainstream models like Anthropic Claude and OpenAI Codex, but not major gains in offensive capability.

The findings suggest current AI guardrails are limiting abuse, and that AI’s impact on cybercrime may be smaller than feared and mostly low-complexity.