What happened
On Sept 29, 2026, TrueFoundry announced Onyx AI Guard is available as a custom guardrail on its AI Gateway: enterprises add config-only input/output rails (prompt-injection defense, jailbreak detection, data-exfiltration checks, content policies) evaluated at the gateway on every prompt and response, with ~10ms overhead and OpenTelemetry-traced decisions — no wrapper or app-code changes.
Why it matters
Shows runtime LLM-security enforcement consolidating into AI gateways as a managed, policy-driven layer rather than per-application SDKs — a pattern admins inherit automatically across every model/team routed through the gateway. Useful for orgs already standardizing on an AI gateway.
Applicability
Platform/security teams running TrueFoundry (or similar OpenAI-compatible AI gateways) should enable centralized Onyx-style input/output guardrails for uniform policy coverage as new services onboard.