Onyx Security

No entries match these filters. Clear filters

1. announced Oct 1, 2026 · issue 092

   Onyx AI Guard integrates with Anthropic inference hooks to enforce enterprise AI security policy natively inside Claude Enterprise.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

   • Connects Onyx AI Guard to Anthropic Claude Enterprise inference hooks, so every prompt, tool-call response, and uploaded-file text is evaluated against enterprise AI security policies before reaching the model.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
   • Returns a real-time allow-or-deny verdict at the inference boundary with no agent, proxy, or per-device installation required — coverage extends across Claude, Claude Code, Cowork, and connected tools from a single integration.
   • Enforces Onyx runtime policies to block sensitive or regulated data exposure, credentials and secrets, intellectual property, prompt injectionprompt injectionAn attack where text the model reads, such as a web page, file or tool result, carries instructions that override the user's. It is the main security risk for agents with tools. and jailbreak attempts, and risky content from agent tool results or connector responses.
   • Extends Onyx's Secure AI Control Plane into Claude's native inference path, so the same policy layer governing copilots, coding agents, SaaSSaaSSoftware delivered over the internet and run on the provider's infrastructure, so the buyer pays a subscription instead of installing or operating anything locally. AI applications, MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. servers, and AI gateways also governs Claude without a separate Claude-specific policy model.
   • Supports agent-aware enforcement beyond prompt inspection — evaluates agent identity, permissions, session context, and the action being attempted at the tool-call boundary, enabling Onyx to block destructive operations, mask sensitive information, and restrict tools by policy.
   • Connects Onyx AI Guard to Anthropic Claude Enterprise inference hooks, so every prompt, tool-call response, and uploaded-file text is evaluated against enterprise AI security policies before reaching the model.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
   • Returns a real-time allow-or-deny verdict at the inference boundary with no agent, proxy, or per-device installation required — coverage extends across Claude, Claude Code, Cowork, and connected tools from a single integration.
   • Extends Onyx's Secure AI Control Plane into Claude's native inference path, so the same policy layer governing copilots, coding agents, SaaSSaaSSoftware delivered over the internet and run on the provider's infrastructure, so the buyer pays a subscription instead of installing or operating anything locally. AI applications, MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. servers, and AI gateways also governs Claude without a separate Claude-specific policy model.

   2 more in Everything

2. announced Sep 29, 2026 · issue 090

   Onyx integrates with TrueFoundry AI Gateway via an open-source wrapper, screening every LLMLLMA large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. prompt and response against a single Onyx policy.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

   └──▷ TRY IT
   Deploy the open-source guardrail wrapper so TrueFoundry's AI Gateway can call Onyx to screen every prompt and response.
   copy

   $ git clone https://github.com/truefoundry/integrations-custom-guardrails
   cd integrations-custom-guardrails/integrations/onyx
   cp .env.example .env
   # Set ONYX_API_KEY and WRAPPER_API_KEY in .env
   docker build -t onyx-guardrails-tfy .
   docker run --rm -p 8000:8000 --env-file .env onyx-guardrails-tfy

   Test Onyx guardrail enforcement on a single request without changing a model's default settings, by passing guardrail groups inline.
   json copy

   {
     "llm_input_guardrails": ["onyx-guard/onyx-input"],
     "llm_output_guardrails": ["onyx-guard/onyx-output"]
   }

   • Adds a TrueFoundry AI Gateway integration via an open-source guardrail wrapper that connects Onyx's evaluate API to TrueFoundry's custom-guardrail contract, registering /onyx-input and /onyx-output endpoints to screen prompts and responses.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
   • Supports per-request guardrail attachment using llm_input_guardrails and llm_output_guardrails headers, so teams can test enforcement without modifying a model's default settings.
   • Blocked requests return the policy message configured in the Onyx console, with no wrapper redeploy required when enforcement rules change.
   • Guardrail decisions are traced as OpenTelemetryOpenTelemetryAn open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. spans alongside latency and cost data, visible in existing stacks such as Grafana, Datadog, or Prometheus.

   4 details in Everything