HiddenLayer, the leading AI security company, announced Agent Harness Security, a new solution that extends the HiddenLayer AI Security Platform's Runtime Security module to secure AI coding agents at runtime. The new solution gives security teams the runtime control they need to safely scale AI-assisted software development by protecting source code, secrets, and infrastructure from unsafe agent actions without slowing developers down. Built on HiddenLayer's AI Security Platform and powered by HiddenLayer's adversarial AI threat research, Agent Harness Security integrates directly into each coding agent's native hook surface to detect and stop prompt injection, sensitive data exposure, and unsafe command execution as they occur.

AI coding agents are rapidly becoming central to enterprise software development. Gartner predicts that by 2028, 90% of enterprise software engineers will use AI code assistants, up from less than 14% in early 2024. But these tools have moved far beyond mere code suggestions. They can now run shell commands, modify repositories, install dependencies, and act on context pulled from files, tool outputs, and external sources.

That shift turns every developer's machine into an AI execution environment, where prompt injection embedded in a README or a poisoned tool response can turn legitimate developer access into the attack path. Organizations cannot safely scale AI coding agents unless they can see what those agents are doing and stop unsafe actions when needed. They also cannot rely on controls that sit outside the workflow and slow developers down, breaking the productivity gains AI coding tools were meant to deliver.

Agent harnesses are the execution environments that enable AI agents to plan, reason, and take actions by connecting models with context, memory, tools, skills, and orchestration logic. As organizations increasingly adopt AI agents, these harnesses create a new runtime attack surface where indirect attacks can influence agent behavior and lead to data exposure, unauthorized actions, or other security risks.

For AI coding agents, Agent Harness Security provides inline runtime control within the coding agent workflow. It gives security teams visibility into agent activity across prompts, tool calls, shell commands, file edits, and repository interactions. It can redact sensitive information before a model sees it, steer agents away from poisoned tool responses, and stop unsafe actions before they pose a risk. Because every coding agent exposes different controls, HiddenLayer applies the strongest enforcement each platform supports and clearly shows security teams what is working, where it is working, and why.

"AI coding agents are now among the most active AI systems inside the enterprise. They run on developer machines, execute commands, modify source code, and act on sensitive context in real time," said Chris Sestito, CEO of HiddenLayer. "Securing them takes more than deciding whether to allow or block an action. Security teams need to control what agents see, reason over, and act on before something goes wrong. That is what our Agent Harness Security solution delivers, securing AI coding agents today, and it's what enterprises need to scale AI coding agents with confidence."

With Agent Harness Security, organizations can:

• See what AI coding agents are doing in the codebase. Follow coding agent activity from the prompt that started the task through the files opened, commands run, dependencies installed, code changed, and pull requests prepared. Security and AppSec teams get the session context they need to quickly investigate behavior.
• Catch threats inside the development workflow. Detect prompt injection embedded in source files and tool outputs, secrets flowing into AI tool calls, unsafe command execution, malicious or unexpected dependency installs, and obfuscated payloads designed to evade review.
• Keep developers moving while reducing risk. Redact secrets before they reach the model and steer agents away from poisoned tool responses with corrective context. Where block-only enforcement would interrupt a long-running CI/CD pipeline and force a developer to step in, content-shaping lets the agent continue safely on its task and preserves the productivity gains AI coding agents were brought in to deliver.
• Trust what is actually enforced. Get clear reporting on which controls are active for each coding agent and the enforcement level supported by each platform. Security teams can see whether an action was detected, redacted, blocked, or limited by the underlying agent platform.

"As we scale the use of AI coding agents across our engineering organization, maintaining visibility and control is essential. Our focus is not only on enabling engineers to benefit from AI, but also on empowering citizen developers to experiment with confidence — all while ensuring agent activity can be monitored, governed, and constrained when necessary. That's what gives us the confidence to move fast in a space where security and compliance can't be an afterthought." - Jaclyn Miller, CPO & Vanessa DeGennaro, VP of Engineering

HiddenLayer's Agent Harness Security solution is available today as a stand-alone solution, securing AI coding agents and extending HiddenLayer's runtime protection directly into developer environments where AI agents read code, execute commands, and interact with enterprise systems.