AI Lab — research, red-teaming and frontier experiments
Experiments ahead
of the rulebook.
Ready when you are.
Adversarial testing, prototypes and proofs of concept, framework evaluation and applied research — published where the work is ready to share.
Enterprise AI Lab
Where we try what’s next — before you have to decide about it.
The Enterprise AI Lab is EAIC’s research division. We use it to stay ahead of the threat and compliance landscape — running adversarial testing on AI systems, evaluating emerging governance frameworks before they become mandatory, and publishing what we find where the work is ready to share.
Current work includes red-teaming methodology for high-risk AI systems under Article 9 of the EU AI Act, evaluation of ISO 42001 certification pathways for mid-market regulated firms, and a risk-adjusted ROI framework for agentic AI systems that sits outside existing governance models.
Research outputs are published on the research page. Organisations can commission the Lab directly for applied research and red-teaming engagements. The build programmes themselves live on the members’ floor — see the dossier wall.
Three ways into the Lab
Where it points, what it publishes, and who it built with.
The Lab is how we enter a sector, and the reason our advice is ahead of the rulebook rather than behind it. These three pages are the output a visitor can actually use.
Where it points
Industries
Four sectors chosen by consequence, not by budget — legal and financial systems, food and local resilience, carbon and waste, and healthcare as analysis only. One learned properly at a time.
See the four sectors
What it publishes
Insights
Board briefings, implementation guides, published research and the 12-question governance scorecard. Working methods rather than thought leadership — written to be used by someone else’s team.
Read the library
Who it built with
Our Work
The named client organisations, the regulated ones who stay anonymous by design, and the dossier wall — six prototypes documented as they run, with their real statuses.
See the client work
How the Lab feeds the firm
Everything starts as a question we cannot answer.
Work enters the Lab as a question, gets tested until there is a defensible answer, and then either opens a sector, goes into something we sell, or stays open. Most of it stays open, and that is the point — a lab whose every experiment succeeded was not running experiments. This is what the dossier wall is: not a portfolio, a pipeline.
| Dossier | The question | Where it points | Status |
|---|---|---|---|
| Carbon Ledger | Can a carbon claim be recorded so it survives challenge years later? | Carbon & waste | opened this sector |
| Evidence Graph | What if governance evidence were a graph instead of a filing cabinet? | Financial systems | still open |
| Fingerprint | What would an outsider find out about you before you noticed? | Financial systems | still open |
| Panopticon | How much can an environment know about itself without knowing who you are? | Next — spatial | still open |
| Rentarobot | Should hiring a robot be as easy as hiring a van? | Next — robotics | still open |
| Arbitrage Engine | Can six agents make decisions at machine speed, with someone still answerable for them? | Next — agentic control | still open |
“Where it points” is the sector a dossier is relevant to, not a claim that it has shipped there. One has opened a sector so far. The rest are open questions, and we would rather say which is which.
How the Lab works
What is on the bench, and what we publish about it.
Each of these is published openly and available as a commissioned engagement. The method is free; the work is not.
Red-teaming
Attacking a system on purpose, before someone else does it for free. Adversarial testing frameworks for high-risk AI systems under EU AI Act Article 9 obligations — establishing what “adequate testing” means in practice. The Lab publishes what it finds; the bought version is a paid engagement.
Agentic AI governance
Governance and oversight models for autonomous AI agents that operate outside the human-in-the-loop frameworks current regulations assume — plus risk-adjusted ROI modelling for agentic deployments.
ISO 42001 certification pathways
Mapping ISO 42001 requirements against the Sentinel governance programme, and building an evidence bundle that supports certification without duplicating work a Sentinel engagement has already done.
Shadow AI detection
Evaluation and improvement of AutoDiscover methodology — detection rates across cloud platforms, API gateways and OAuth permission graphs, and published findings on shadow AI prevalence in regulated estates.
Model evaluation
Evaluating whether a model does what someone claims, in the conditions it will actually meet. The Lab publishes the method; the bought version of this work is a paid engagement.
Robotics and spatial intelligence
The governance problems arrive before the robots do: who may operate one, who is accountable when it acts, what it is allowed to record, and what evidence survives afterwards.
The members’ floor
The dossier wall.
The Lab builds with client partners, not for spectators. These are the declassified covers of what is on the bench — the working notes, prototypes and build programmes sit behind an engagement. More dossiers are declassified as projects leave private beta.
Dossier 01 — Robotics & spatial intelligence
RENTAROBOT
Hiring a robot should be as easy as hiring a van.
Booking, identity, insurance evidence, teleoperation and an operator marketplace for physical machines. The hard part isn’t mechanical — it’s who answers when nobody’s driving.
Read the dossier →Dossier 02 — Spatial intelligence — experimental
PANOPTICON
How much can an environment know about itself?
Total situational awareness for physical environments — splat-captured digital twins with change detection, governed hard enough to earn the name.
Read the dossier →Dossier 03 — AI & intelligence
ARBITRAGE ENGINE
Six agents. Machine-speed decisions. The governance problem of who answers for them.
Scout, Analyst, Executor, Guardian, Orchestrator, Historian — an autonomous trading architecture where Guardian answers the regulator and Historian answers the court.
Read the dossier →Dossier 04 — Circular systems
CARBON LEDGER
Carbon claims that survive scrutiny.
Every tonne traceable to a collection, a batch record and a process run. Carbon evidence built for audit, not marketing.
Read the dossier →Dossier 05 — AI & intelligence — core research
EVIDENCE GRAPH
Governance stops being a filing cabinet when evidence becomes a graph.
One graph across entities, systems, AI, assets, contracts, decisions and risk — so governance questions have answers instead of meetings.
Read the dossier →Dossier 06 — AI forensics
FINGERPRINT
What would an outsider find?
External attack-surface discovery and AI fingerprinting — detecting AI where nobody declared it.
Read the dossier →Working notes, prototypes and lab time are available to client partners under NDA. The Feasibility Sprint is the usual way in; a Lab Retainer buys standing capacity.
Published research
What we have published.
Working methods rather than thought leadership — written to be used by someone else’s team, and to be criticised by someone else’s regulator.
The unassigned workstream
Councils are being abolished to a statutory deadline and nobody owns the AI workflows, evidence or audit records. Five tests and a handover schedule.
Read the noteRed-teaming high-risk AI under Article 9
What EU AI Act Article 9 implies for adversarial testing — and the staged method the Lab uses, from threat modelling to evidence capture.
Read the noteGoverning agentic AI: the open questions
Why autonomous agents break human-in-the-loop assumptions — the concrete governance gaps, and the Lab’s working positions on interim controls.
Read the noteThe Inception Hack: nested prompt injection
The most dangerous prompt is the one nobody wrote. How layered AI memory systems can be turned against their owners — and what Sentinel looks for.
Read the paperHave a question the rulebook hasn’t caught up with?
The Lab takes commissioned engagements — red-teaming high-risk systems, experimental prototypes and proofs of concept, evaluating emerging frameworks, and applied research for organisations facing questions ahead of the rulebook.
Scoped per engagement · NDA standard · Lab access is for client partners