Attackers Target AI Coding Tools to Steal Credentials, Report Finds
- Severity
- High
- Sources
- 1
What happened
Recorded Future's Insikt Group reports 103 AI-linked incidents from 31 July 2025 to 31 July 2026, of which 32 involved AI in malware or offensive operations and nine involved state-sponsored collection. Late February 2026 saw compromise of a popular AI proxy package delivering credential theft, lateral movement and backdoor persistence, and April saw AI tool packages used to harvest credentials and access crypto wallets. Shai-Hulud expanded to target Claude Code in May, and malicious AI coding plugins forwarded provider API keys to attacker infrastructure. Defenders should separate coding agents from sensitive systems, require approval for consequential actions, audit AI dependencies, plugins and model repositories, and monitor exposed API keys, OAuth tokens and stolen AI subscriptions. No hashes, domains or IPs were published.
For security analysts · Written by AI from 1 source · Wrong?
What to do
- Separate coding agents from sensitive systems and restrict what they can access and change.
- Require human approval for consequential actions taken by AI agents.
- Audit AI dependencies, plugins and model repositories for backdoors or malicious command injection.
- Monitor exposed developer API keys, corporate OAuth tokens and AI subscriptions for misuse.
Written by AI from the sources, checked against them
Full story timeline · 1 update
Who reported · 1 outlet
Who and what is in it
Malware
Companies
Products