AI changes the risk picture
AI systems introduce risks traditional security reviews weren’t built for. A language model can be manipulated by text hidden in a document it reads. An agent with broad permissions can be talked into actions nobody intended. Prompts and outputs may contain personal data that ends up in a provider’s logs. Regulators are paying attention: data protection law already applies to AI processing, and the EU AI Act adds obligations that depend on how and where AI is used.
Security, designed in
We approach AI security the way good engineers approach any system: understand the data flows, identify what could go wrong, and put proportionate controls in place. Our reviews draw on the OWASP Top 10 for LLM Applications and the NIST AI Risk Management Framework, covering prompt injection, sensitive information disclosure, excessive agency, insecure output handling and supply-chain risks. The fixes are concrete: least-privilege tools, confirmation for sensitive actions, separation of trusted and untrusted content, PII redaction, output validation and complete audit logs.
Data protection by design
For GDPR and similar laws, we map what personal data each AI system processes, for what purpose, where it goes and how long it is kept. We help minimise data before it reaches a model, choose providers and settings that don’t train on your data, configure retention and support data subject requests. The result is documentation your DPO can work with.
Regulatory awareness, not guesswork
We help you classify AI use cases against the EU AI Act’s risk categories and identify which obligations might apply — transparency for chatbots, stricter requirements for high-risk areas such as recruitment or education — so your legal advisers can confirm and you can plan. We’re careful to be clear about what is engineering and what needs a lawyer.