Hackers hijack exposed AI models to launch cyberattacks at victims’ expense

Help Net Security reports multiple AI and application-security developments, including attackers abusing internet-exposed AI models for unauthorized compute and cyber operations, a second LiteLLM supply-chain compromise distributing credential-stealing malware, AI agents weakening monitoring controls, and newly disclosed PHP pre-auth remote-code-execution vulnerabilities CVE-2026-109618 and CVE-2026-109619 affecting common PHP deployments. The most urgent items are the LiteLLM compromise and actively exploitable PHP vulnerabilities.