Assure · Offensive Security & Assurance · Fixed-scope assessment
AI & LLM Application Security Testing
AI applications fail in ways traditional testing does not look for. A crafted message can make a chatbot ignore its instructions, leak data it can see, or push an agent into actions nobody intended. The NCSC has warned that prompt injection may never be fully solved, so the aim is to find these paths and limit what they can reach.
Talk to us about this
Who it is for
Organisations putting large language models in front of customers or staff: chatbots, search over internal documents, and AI agents that can take actions.
What is included
- Prompt injection, both direct and hidden in documents, emails and web content
- Data leakage through retrieval and connected data sources
- Agent and tool abuse: unintended actions, excessive permissions, unsafe integrations
- Guardrail and content-filter bypass
- Review of the surrounding cloud architecture, identity and logging
What you get
- Findings mapped to the OWASP Top 10 for LLM and agentic applications and to MITRE ATLAS
- Reproducible test cases
- Architectural fixes that limit the blast radius, not just prompt tweaks
- A debrief with your product and engineering teams
Testing runs under written authorisation and an agreed scope, against systems you own or are permitted to test.
Why Cipherer
- Principal-led. The person who scopes the work does the work, with no hand-off to a junior team.
- SC-cleared principal consultant (National Security Vetting), and a Cyber Essentials certified company.
- Seventeen years of delivery across government, healthcare, banking and Web3, including an organisation-wide cyber security programme for the Scottish Government.
- Security by design in everything we build and run, from a community website to a bank's data platform.
- One partner from advice to operation, based in Scotland and working across the UK.
Every engagement starts with a short call to agree scope, then a written proposal setting out deliverables, timeline and price.