Skip to content

Back to the catalog

Attackers Didn’t Wait for AI. They Built Workflows Around It.
huntress.comarticle

Attackers Didn’t Wait for AI. They Built Workflows Around It.

Huntress details three real-world patterns where attackers build workflows around AI adoption rather than attacking AI models themselves: SEO-poisoned/malvertised fake "Claude Code" downloads that deliver credential-stealing malware, an AMOS stealer campaign that spoofs ChatGPT/Grok-style answer pages to trick macOS users into running malicious Terminal commands (ClickFix-style), and the EvilTokens phishing-as-a-service platform, which used Railway infrastructure for on-demand session-token theft and AI-generated, role-tailored phishing lures that bypass email filtering. The piece argues AI isn't the weapon but the accelerant, making malicious activity blend into trusted, expected workflows.

TL;DR Huntress breaks down real incidents where attackers weaponize trust in AI tools — fake Claude Code malvertising, spoofed ChatGPT/Grok pages delivering AMOS stealer via ClickFix-style commands, and the EvilTokens phishing-as-a-service platform on Railway using AI to tailor lures and bypass MFA.
Open original ↗

Organisations named here

  • ClickFix
  • John Hammond

Technologies named here

  • Railway
  • macOS
  • Claude Code
  • ChatGPT
  • Grok

Techniques named here

  • Adversarial poetry prompt injection
  • SEO poisoning / malvertising
  • Fake AI-generated instructional pages
  • AI-tailored phishing lure generation
  • Session token theft / device code phishing
  • AI-driven signal correlation for social engineering

Threat actors named here

  • EvilTokens
  • AMOS Stealer

Related in Threat Intel