Vulnerable AI Tools Exposed Online: Langflow, LiteLLM Risks

Publicly accessible AI tools are growing rapidly, with over 294,000 IP addresses associated with AI services detected in early 2026. This is up from 183,000 in October 2025. What's interesting is that the most vulnerable AI products are also the ones appearing online most frequently. For example, the AI agent-building tool Langflow has seen a 169% increase in internet-exposed instances over the past nine months. This is despite having 18 vulnerabilities since 2024, with 14 of them high-severity. Another common AI tool, LiteLLM, has nearly doubled its internet-exposed instances. Hackers are actively exploiting a pre-authentication SQL injection vulnerability in LiteLLM. A compromise here could expose a customer's API keys for various large language model services. The bottom line is that these trends create a vast array of targets for hackers, potentially impacting critical infrastructure and the AI tools companies rely on.