“I’m Allowed”: Hackers Use Simple Claims to Bypass AI Guardrails

Cisco Talos analysis finds cybercriminals abusing AI coding assistants and chatbots to build and operate attack tools, automate large-scale criminal operations (including an email-validation system handling tens of millions of records), perform credential and source-code harvesting across thousands of hosts, deploy autonomous agents that exfiltrated Telegram profiles and wallet records, and manage Monero-mining at scale; attackers often bypass AI guardrails by claiming authorization, and outcomes depend on operator skill.