Cybercriminals are defeating safety controls on commercial artificial intelligence tools by breaking malicious projects into small fragments across multiple sessions. Research published by Cisco Talos analysed prompt logs from attackers using AI coding assistants such as Claude Code, Codex, Cursor and Gemini. Attackers also bypassed guardrails by claiming to own targeted infrastructure or labelling their work as bug bounty activity. Cisco Talos found that existing safeguards provided little protection across platforms, with an operator's existing skills determining what the technology delivered.

Background: Generative AI in Law Enforcement

← Back to the news feed