TORONTO, Oct. 18, 2026 — Check Point Research's disclosure this week of two vulnerabilities in LiteLLM — the open-source LLM gateway with roughly 97 million downloads a month and 59,000 GitHub stars — is a wake-up call for every enterprise routing AI traffic through a gateway layer.

The flaws, tracked as CVE-2026-35029 and CVE-2026-59823, both target the gateway control plane. The high-severity CVE-2026-35029 is a broken access-control issue: an account with the lowest privilege level, such as a read-only user, can reach administrative functions, read server files, overwrite admin credentials, and execute code — effectively taking over the gateway. The medium-severity CVE-2026-59823 is a server-side request forgery (SSRF) flaw that lets attackers redirect gateway requests to their own servers and capture AI provider API keys.

Check Point notes the exposure may be wider than it looks: LiteLLM is embedded in many AI platforms and internal tools, so vulnerable deployments may be running where they were never deliberately installed. Both issues are fixed in LiteLLM 1.83.9, and security teams are advised to check for indirect installations.

"As organizations increasingly rely on gateways such as LiteLLM to connect applications, users and AI services, these platforms become attractive targets because they sit at the centre of trust," said Mateo Rojas-Carulla, Head of Research, AI Agent Security at Check Point Research. "The future of AI security will depend on securing every layer of the AI stack, because attackers will naturally target the one that gives them the greatest access with the least resistance."

The disclosure underscores a simple architectural point: the gateway through which every prompt and API key flows deserves the same security discipline as any other piece of critical infrastructure — least-privilege access to the control plane, inspection of every request and response, strict API-key hygiene, and a full audit trail.

That is the lane Securetron's AI-FW was built for. AI-FW is a self-hosted AI firewall and governance gateway that sits between an organization's AI agents and the models they call. Every prompt and response is scanned in real time for jailbreaks, prompt injection, PII, toxicity, and data exfiltration — inbound and outbound. Semantic intent analysis scores requests against natural-language policies, and rolling risk scores per user, agent, and IP trigger automatic blocking. Every transaction lands in a metadata-only audit log; content is never persisted.

Directly relevant to this week's news, AI-FW's key-management model addresses the failure mode the LiteLLM SSRF flaw exposed: provider keys are stored server-side in the gateway's model registry and injected on forward, so endpoint agents never hold them — shrinking the blast radius if an agent or account is compromised. The gateway is fail-closed by design; agents authenticate via API keys, JWT, mTLS, or Kerberos with RBAC enforcement; and intelligent model routing adds per-model keys and failover.

"The gateway is the new front door of enterprise AI," said Sarah Reid of Securetron. "This week's disclosure shows what happens when that front door's own locks are weak. AI traffic deserves a security layer that is fail-closed, inspects every request, and keeps provider keys off endpoints — that is what AI-FW delivers, self-hosted, in minutes."

AI-FW's Community Edition is free — 100 million tokens per month — for teams that want a governed AI pipeline quickly. (Community Edition is proprietary free-to-use software; it is not open source.) Teams can point any OpenAI-compatible or Anthropic client at the gateway with no agent-side changes, and inspect, route, and audit their first prompt the same day.

About Securetron: Securetron builds self-hosted security infrastructure for the AI era, including AI-FW (AI firewall and governance gateway), PKI Trust Manager (certificate lifecycle management), and Boltron-AI (AI agent for PKI operations).

Press contact: Sarah Reid, Securetron, 1 Yonge St, Toronto, ON M5E 1E6, Canada, +1 (438) 927-8036, [email protected]