Co-Intelligence Security
The people–AI collaboration framework
Multiple professionals and AI agents, each bringing their strengths, investigate, analyse, develop and verify together.
People bring their expertise to the whole process, from defining the problem to verifying and improving the result.
How do people and AI actually work together?
Several people and AI agents take different roles and work together; permissions, approvals and records are handled in one layer above the work.
Reviewed
The parts that work together
A task draws on these parts together. They work towards the same goals in one workflow, and what is learned along the way builds up as security capability.
- People and organisations
Security experts who define problems, form hypotheses, experiment, analyse, build and judge. Your development, operations, audit and decision-making staff and our research and education partners take part in their own roles. Roles are divided by the difficulty of the task, the information and permissions involved, and what each participant can do: people also carry out the work themselves, and AI agents also take on analysis and development.
- Specialist tools
Tools for attack, assessment, analysis and validation, used directly by people and also by AI agents.
- AI models and security agents
The external, internal and local models agreed and connected for each task, and role-specific security agents equipped with tools, knowledge, verification and collaboration.
- Shared coordination
One place that brings together the work split between people and agents, its status, context and results, evidence and approvals, so everyone can see who is doing what and it stays under control.
- Data, knowledge and playbooks
The expertise and experience, attack and defence techniques, playbooks, evaluation datasets and quality standards delivered to people, AI agents and models for use in a task.
- Operation, control and verification
Roles and permissions, approval and stop procedures, reasoning and evidence, quality and safety evaluation, and policy compliance.
- Training and evaluation
Hands-on and scenario training, controlled labs and challenges for people, and evaluation and improvement of AI agents' task performance and safe operation.
- Services and partners
Services that combine experts, tools and AI, working alongside partners' technology and data.
Six steps from connect to improve, repeated as needed
We design how strong models, specialist tools, several AI agents and experienced people work together across six steps, aiming for results none of them could reach alone.
Connect
We connect the models, tools and data agreed for each engagement into one workflow.
Coordinate
Defining the problem, hypotheses and experiments, investigation, analysis and building are split between people and AI agents, and context and results pass through one place.
Observe
Each step is recorded: who did what, with which tool, and how far.
Control
Anything outside the set permissions is blocked, important actions need a person's approval, and work stops when an agreed stop condition is met.
Verify
Whether people, tools or AI found it, a finding is cross-checked and becomes a result only after an expert has filtered out false positives, and it is kept with its evidence. Where more is needed, the work goes back a step.
Improve
Within the permissions and scope agreed, reviewed methods and knowledge go into tools, playbooks and the way knowledge is delivered, and are used again in the next engagement.
Where the models available for a task are limited, we work in the same flow: splitting the work, running tools, and adding shared knowledge, cross-verification and human judgement. Why this matters: Sovereign Security AI
Where this leads
Why and how we build security capability that stays with the organisation as models change
Six research domains and their main research topics
This page describes the collaboration framework we research and adapt to each engagement.
People & AI Collaboration (COLLAB) — the domain that researches the technology of people and AI agents working together. This framework draws on its results and on those of other domains such as KNOWLEDGE, MODEL and AGENT.