Onyx Security

A cyber incident can start with an authorized AI agent doing exactly what it has permission to do. No attacker required.

Our Chief Strategy Officer, @youngbloodtim, joined Krista Case from @theCUBE to discuss unintended actions and rogue agents in the enterprise.

A few points that stood out:

1️⃣ Runtime control is essential. You need to understand intent and agent reasoning in order to intervene in real time.

2️⃣ Identity needs context. Whose authority is an agent using? Do its permissions fit the task? Who is accountable when it acts?

3️⃣ An AI kill switch needs nuance. Steering an agent back on task, limiting its permissions, and quarantining it are different responses to different levels of risk.

4️⃣ Agent failures belong in your tabletop exercises. Test your team’s ability to detect, contain, and recover from unintended agent actions while keeping the business running.

Tim leaves security leaders with a practical test: could you stop an unsafe action before it runs?

Watch the full conversation: youtube.com/watch?v=ab2eHh…

Onyx has been named to the Fortune Cyber 60! 🎉

@FortuneMagazine, @lightspeedvp, and @AWS evaluated over 300 venture-backed companies and awarded Onyx a place in the early-growth-stage category.

The recognition comes as enterprises give AI agents greater responsibility - and face growing pressure to secure them. A Wakefield Research CISO survey accompanying the report found:

➡️ 88% of companies already allow AI to take some response actions without a human in the loop.

➡️ 79% experienced a confirmed or suspected AI-related security incident in the past year.

Helping enterprises navigate that reality is why we built the Onyx Secure AI Control Plane: to give teams the visibility, governance, and runtime protection to put advanced AI to work with confidence.

Thank you to our customers, partners, investors, and the Onyx team for helping bring our mission to life.

Read our announcement: onyx.security/blog/onyx-name…

Onyx is a Gartner Cool Vendor in Agentic AI Security! 😎

This recognition underscores the importance of our mission to control advanced AI, which is increasingly urgent as models and agents become more capable and autonomous.

Enterprises need to protect against adversarial threats, employee misuse, and rogue agents everywhere AI is used. This requires an understanding of intent, permissions, and agent reasoning.

The Onyx Secure AI Control Plane brings that context to runtime decisions. It can block, steer, mask, or escalate an action for human review, giving teams more ways to keep work within authorized boundaries while helping legitimate workflows continue.

We think that's pretty cool. 🤘

To learn more, check out our blog: onyx.security/blog/onyx-name…

We’re bringing Onyx runtime security natively into Claude through our integration with @AnthropicAI inference hooks, enabling companies to protect their entire Claude environment on any device. 🔌

The integration extends the Onyx Secure AI Control Plane directly into Claude Enterprise, evaluating prompts, tool-call responses, and uploaded file content against enterprise security policies before they reach the model.

Companies can block sensitive data exposure, malicious requests, and unauthorized use on any device, with nothing to install - extending to contractors and other third-parties who don't have company issued devices.

As Claude moves from answering questions to performing work, the security challenge grows. A legitimate request to investigate a code issue can lead an agent to access production secrets, modify configurations, or send information to an external service. Securing that workflow requires understanding the agent’s permissions, intent, and actions alongside the content it processes.

Anthropic inference hooks add a native checkpoint to Onyx’s broader runtime protection, connecting Claude to the same policy layer enterprises use to govern and control AI across endpoints, browsers, SaaS, cloud, and code.

Deep native enforcement. One Secure AI Control Plane across enterprise AI.

Available now for organizations using Onyx and Claude Enterprise.

Check out our blog: onyx.security/blog/onyx-inte…

Onyx is excited to announce an integration with the @TrueFoundry AI Gateway.

AI gateways are an important component of the Onyx Secure AI Control Plane. We're pleased to extend support to a preferred gateway solution of Fortune 1000 companies, TrueFoundry.

With Onyx runtime protection within the TrueFoundry AI Gateway, enterprises can now enforce security policies across every LLM prompt and response from a single place.

⛔️ No more per-application integrations.
⛔️ No policy drift.
⛔️ No gaps when a new AI service ships.
🟢 Just centralized protection against prompt injection, jailbreaks, data exfiltration, and policy violations across the models your teams use.

One gateway. One policy layer. Every LLM call protected.

Read more about the integration here: onyx.security/blog/onyx-and-…