CyberLab Among the First Four Companies Globally to Achieve CREST's Security Testing of AI Accreditation
CyberLab has become one of the first four companies in the world to achieve CREST’s new Security Testing of AI accreditation, demonstrating independently assessed capability to test the security of Generative AI (GenAI) and Large Language Model (LLM) enabled systems.
CREST, the global not-for-profit body that accredits cyber security providers, opened the accreditation for applications in August 2026. Developed with CREST’s AI Working Group, it establishes independently assessable requirements covering technical expertise, testing methodology, governance and quality controls. It is a supplement to CREST’s established Penetration Testing Accreditation, which providers must already hold, so it builds on capability that has already been independently assessed rather than standing on its own.
The accreditation comes as AI moves quickly into day-to-day business use. Office for National Statistics data shows 29 per cent of UK businesses were using at least one AI technology by June 2026, rising to 49 per cent among organisations with 250 or more employees. For organisations buying AI security testing, CREST accreditation means a provider’s capability has been checked independently, rather than taken on trust.
Our customers are putting AI in front of their own customers, often faster than their security teams can keep pace with. They need to know it won’t leak data or do something it shouldn’t, and they need to know the people testing it are qualified to judge. We’ve been testing AI systems for clients for some time, so having CREST assess that work independently is a proud moment for our testing team. They’ve earned it.
– Gavin Wood, CEO
New AI Security Testing Service
Alongside the accreditation, CyberLab has launched a dedicated AI Security Testing service for organisations building on large language models. Typical projects include customer service chatbots, copilots built into a company’s own software, AI agents that can take actions in other systems, and tools that answer questions using a company’s own documents, known as retrieval-augmented generation, or RAG.
Each test covers the model, the application built around it, the data it draws on and the tools it can reach. Testers look for weaknesses including prompt injection, where an attacker slips hidden instructions into content the AI reads; leaks of sensitive information; AI agents with more access than they need; and denial of wallet attacks, designed to run up a company’s AI usage costs.
Testing follows the OWASP Top 10 for LLM Applications and supports organisations working towards ETSI EN 304 223, the European standard for AI cyber security, which builds on the UK Government’s AI Cyber Security Code of Practice. Rate and cost limits are agreed with the client in advance, so testing can run safely on live systems as well as in test environments. Where an application relies on a third-party model provider, testing stays within that provider’s terms of service.
What CyberLab’s AI Security Testing Covers
The service assesses the whole AI system, from the model and the application around it to the data it retrieves and the tools it can reach.
- Systems tested: Chatbots and conversational interfaces, copilots embedded in a company’s own products, AI agents that call tools and act on a user’s behalf, retrieval-augmented generation (RAG) pipelines and vector stores, and custom or fine-tuned models.
- Risks assessed: Prompt injection, sensitive information disclosure, excessive agency, improper output handling, data and model poisoning, and denial of wallet.
- Standards applied: Aligned to the OWASP Top 10 for LLM Applications, and supporting the provisions in ETSI EN 304 223, the European Standard for AI cyber security developed from the UK Government’s AI Cyber Security Code of Practice.
- Hosted models included: Applications built on providers such as OpenAI, Anthropic and Azure OpenAI Service are tested within the provider’s terms of service, agreed in advance.
- Safe on live systems: Testing runs under strict rate and cost controls, so it can be conducted on production systems with users present, and on staging environments where a client prefers.
Every engagement closes with a jargon-free report covering each finding, its potential impact, a severity rating and specific remediation guidance, with consultants available to walk the client’s team through it.
Most of the risk in an AI system isn’t in the code. It’s in how the system behaves, and traditional pen testing was never designed to find that. CREST’s accreditation means clients don’t just have to take our word that we can test it properly.
– Steve Clarke, Head of Penetration Testing
CyberLab was named in CREST’s announcement of the first four accredited companies, published on 3 November 2026.
The accreditation follows CyberLab’s recognition as a Founding Signatory of the CREST AI Charter and builds on the company’s existing CREST and CHECK accreditations for penetration testing and red teaming.
The AI Security Testing service is available now. Organisations can find out more here.
About CyberLab
CyberLab is the UK-leading cyber security consultancy and managed services provider, trusted by over 1,200 enterprise businesses, government departments, and household names to secure their operations, systems and data. With more than 30 years of combined expertise, we take a deeply consultative, partnership-led approach guiding clients at every stage of their cyber journey. Protecting the nation, business and people.