Cloud Range launches AI readiness framework for firms

Cloud Range has launched its AI Validation Range and AI Readiness Framework for organisations testing whether artificial intelligence systems are ready for operational use.

The launch reflects growing cybersecurity concern over AI agents that can complete assigned tasks while also straying beyond intended limits, including by accessing systems or taking actions that were not anticipated.

The new offering gives companies a contained environment to test AI models and agents in security operations centre scenarios and cyberattacks before giving those tools wider responsibility. It is also designed to compare the performance of AI systems with that of human defenders in specific roles.

The aim is to help security teams decide which tasks AI can handle, which still require human judgment and oversight, and where a combination of the two is most effective.

Testing environment

The AI Validation Range is built on Cloud Range's existing cyber range platform and recreates enterprise environments with licensed tools and generated network traffic. It also draws on a library of automated adversary attack emulations, allowing customers to assess AI behaviour under hostile and unpredictable conditions without touching live production systems.

Cloud Range has positioned the launch as a response to a shift in how AI is being used inside organisations. Rather than limiting systems to recommendation and analysis, some employers are considering AI agents for more autonomous operational work, particularly in security environments where speed and scale matter.

That shift raises a different set of questions from conventional software testing. Beyond whether an AI system can complete a task, security leaders must assess how it behaves under pressure, what it does when faced with unexpected inputs, and whether it stays within defined access and authority boundaries.

"AI is moving from recommending what humans should do to actually doing it, and that fundamentally changes the risk equation," said Debbie Gordon, Chief Executive Officer, Cloud Range.

"Recent events have made clear that a successful test is not the same thing as proven readiness. An AI agent can accomplish its assigned objective and still take a path no one expected or intended. Organisations need to understand not just whether AI works, but how it behaves under pressure, where it fails, when it needs human intervention, and what level of autonomy the evidence actually supports. Organisations cannot afford to find that out in a production environment," Gordon said.

Five-step framework

Alongside the testing platform, Cloud Range introduced an AI Readiness Framework based on what it calls a five-step PROVE process. The framework covers preparation and training, risk assessment, operational testing, validation and benchmarking, and ongoing evaluation as models, threats and workflows change.

The framework was shaped by work with organisations already using the AI Validation Range to assess AI performance under realistic operating conditions. Its stated aim is to move deployment decisions away from assumptions and towards evidence gathered through direct testing.

For cybersecurity teams, benchmarking AI against human analysts could become an important part of that process. Security operations centres are under pressure to handle large alert volumes, investigate incidents quickly and maintain round-the-clock coverage, making them a likely early testing ground for agent-based AI tools.

At the same time, the sector faces clear risks in handing too much authority to systems that may behave in unexpected ways. Access rights, decision-making autonomy and the possibility of unintended actions all become more consequential when AI is allowed to act directly inside production environments.

Cloud Range says its testing environment is intended to let organisations examine those issues in advance, including whether an AI system's role should remain tightly bounded or whether evidence supports broader autonomy.

"Organisations shouldn't discover what an AI agent is capable of accessing, changing, or breaking for the first time in production," Gordon said.

"The goal isn't to prove that AI works. It's to understand how it works, where it performs well, where it doesn't, and what success actually looks like before you give it greater responsibility. AI readiness has to be continuously proven," she added.