Financial institutions
The same failure recurs wherever consequential decisions are automated and capital sits behind them: an AI optimising locally while exposure compounds globally, every individual decision compliant, the aggregate drifting somewhere no one chose. Triodian governs the book, not just the decision, and gives you evidence you can verify.
Every position clears mandate. Your risk report rolls up green. It probably should, each decision was compliant when it was made. That is exactly why the thing that's drifting doesn't show up in it.
Why now
In November 2025, a state-sponsored group ran an autonomous AI agent through a full intrusion campaign against roughly thirty organisations - with an estimated 80% of the operation executed with no human in the loop. The same capability that made that possible is now inside the engines allocating, pricing and underwriting capital: an AI can carry thousands of individually-compliant decisions to completion faster than any quarterly review can reconstruct what they add up to. By the time the aggregate is visible, the exposure is already on the book.
~80%
of the operation ran autonomously, with no human in the loop.
~30
organisations targeted in a single agent-driven campaign.
Where this applies
Investment-signal or allocation models that drift toward concentrated, correlated positions while each trade clears mandate, relevant to trustee diligence and CPS 230 operational-risk obligations over AI-driven processes.
Credit and pricing engines that build correlated exposure or conduct risk across a portfolio while every individual loan stays comfortably inside policy.
Underwriting AI that thins tail-risk margin and stacks correlated peril book-wide, assembling a solvency exposure entirely out of individually compliant policies.
Any target whose value rests on an AI running a consequential path, where the aggregate behaviour is the asset you are buying, and the risk you inherit at completion.
Getting to the data
The first practical objection is access. There are three routes to it, in descending order of ease, and we are candid that the hardest case is not solved by software.
An owner asking an investee to instrument a risk it already carries. This is the natural home of the capability, the incentive and the access already sit on the same side.
Governance becomes a condition of the deal, not a demand a target can refuse, most valuable where deep diligence into AI behaviour was previously uneconomic to attempt.
Answered only by the hardware appliance: the target runs its own decisions through a verifiable instrument, keeps its data, and produces the attestation. We do not pretend software solves this, this route is gated and optional.
What it changes
Evidence about what an AI’s decisions mean in combination, the layer ordinary diligence cannot reconstruct.
Enforcement holds whether or not the team’s skill to supervise the machine stays intact, the last line of defence isn’t a competence you can’t measure.
Compounding tail risk becomes detected and enforced rather than invisible, lowering the risk carried on the balance sheet you’ve acquired.
A tamper-evident record of what was governed converts “we trust the operator’s guardrails” into something an outside party can verify after the fact.
Regulatory context
The obligations are already written. What’s missing is a way to prove, to a regulator or a board, that an AI-driven process actually held to them. The appliance produces exactly that evidence, it sits beneath these frameworks rather than competing with them.
High-risk system obligations on governance, record-keeping and human oversight.
High-risk provisions in force Aug 2026Operational-risk obligations over critical, AI-driven processes and controls.
Operational risk managementThe management-system standard for governing AI across its lifecycle.
AI management systemsThe risk-management framework institutions map their AI controls against.
AI risk management frameworkHow the appliance protects the position
The appliance compares what the AI’s cumulative behaviour actually means against what the institution’s stated risk appetite means, so the drift is flagged as it forms, enforced at the point of decision, and bound to a record an outside party can check.
Drift accumulates inside compliant decisions, undetected by the controls the institution relies on, and carried in full on the balance sheet.
The same risk becomes one that is flagged as it forms, enforced at the point of decision, and backed by a record an outside party can check, lowering the risk you carry once invested.