QUANTENT · THE FUTURE OF ENTERPRISE SAFETY
QuantEnt is the pre-action control layer for the AI enterprise. We know what everyone in your company — human or AI — is allowed to do, we fix what’s wrong, and we make sure AI cannot do anything it shouldn’t. Before the action, not after.
01What we do
Enterprises run on permissions and data, and both are ungoverned. Access accumulates unchecked. Agents act without knowing what words mean. Compliance teams fly blind. And now regulators have written a deadline. Each of these is a separate problem to most vendors. To us they are one problem: nobody knows what anyone is allowed to do, and nothing checks before the action.
Every agent states its intent, gets its meaning resolved, is checked against its mandate and limits, and receives a single-use grant before it acts. No grant, no action. Nothing is guessed. Humans are asked when something is off.
We compile your entitlements and business processes into one computable model, price every permission in dollars, and shrink every footprint to what the role actually needs. The quarterly “Approve All” wall goes away.
Blast radius, toxic combinations, key-man exposure, agent limits and containment — measured, priced, and controlled. You choose the maximum daily damage. We enforce it.
The EU AI Act’s record-keeping, deployer and monitoring duties — Articles 12, 26 and 72 — handled by the same mechanism that prevents the harm. The record is a by-product of the approval, not a project of its own.
02Making AI trustable · QuantData
Financial services want agents. They will not let them act, because nobody can say what an agent may do, or stop it when it does something else. QuantData is the pre-action control plane that answers both. The agent proposes; QuantData governs. No consequential action executes until its meaning is certain, its actor is inside its mandate, its limits hold, and a human has been inserted where policy demands.
Most data failures are not bad numbers in the warehouse. They are the right numbers used in the wrong sense, at the wrong version, at the wrong time. QuantData resolves every request to an approved definition and an official source before a token is spent or a byte is read. “Hedge all accounts” is stopped, the five valid meanings of “account” are surfaced, and nothing runs until the one that was meant is confirmed.
Your action plane accepts nothing without an operational grant QuantData issued: single-use, bound to the decision, hashed to the exact request. The agent someone launched last night without telling anyone has no path to act. Logging tools find that agent in the incident report. We give it nothing to act with.
When a request is ambiguous, out of mandate, over a limit, drifting from its baseline, or assembling a control failure with another agent, the action stops and a named person receives a specific question with the evidence attached. That is how the people who must oversee your AI find out they are needed.
The pre-action control loop
No grant, no action.Agent states what it is trying to do
Resolved to canonical concepts; ambiguity stopped
Checked against the human-assigned operating surface
Hard limits over live telemetry; agent-to-agent policy
Single-use, fidelity-hashed grant
Governed endpoint verifies offline; no grant, no action
Append-only, bitemporal record of intent, verdict, approvals and denials
Assigned by declaration, not code
Every agent, registered or not. Intent recorded with events. Article 12 from day one; shadow agents surface.
Ambiguity dies at the gate. Official sources, version bridges, start-of-day readiness checks. Data quality, enforced.
Registration, charters, out-of-lane denial, approval insertion, aberrance escalation, permission shutoff.
Standing permissions to zero. Per-operation grants, fidelity, brokering. Halt-issuance is the kill switch — zero lag, other agents untouched.
Semantic integrity
Relative spend index. Illustrative. QuantEnt estimate from customer workloads; the retry loop is eliminated before a token is spent.
QuantData sits above your data, not in its path: no bottleneck, no bypass. External agents enter through a single MCP gateway. Works with Claude, OpenAI, Microsoft Copilot and your own agents. SaaS, or on-premises for regulated environments.
03Cleaning up your permissions · QuantCertify + QuantVault
Role sprawl and manual access reviews create audit failures, insider risk and regulatory exposure, and nobody can say what any of it costs. QuantCertify compiles your entitlements and business processes into one computable model. Every permission is mapped to the business step where it is exercised, priced by the damage it could do, and shrunk to the footprint the role actually needs. Say goodbye to audit failures and “Approve All”.
Before
Certification season. A wall of incomprehensible entitlement names lands on the head of product control every quarter. He gives up and clicks Approve All. The control exists and does nothing.
After
No quarterly wall. The moment an employee is suddenly exposed to “PII exposure risk” — in words, not an entitlement code — management is alerted and the fix is in. Certification is continuous and event-driven, not calendar-driven.
Every entitlement’s practical damage potential, in dollars. Smaller footprint, smaller blast radius.
Fires on envelope breach, peer deviation, toxic-combination emergence, threat shift and ten other triggers. The calendar is replaced.
Permissioned authority versus exercised authority. The delta is stale access, unnecessary risk, or misdeployed talent.
Roles inferred from what people are permissioned and what they actually do, not from HR titles. Assigned versus permissioned versus telemetry versus capability.
Every user against their true peer cluster and their own baseline. “Unusual for someone like them” becomes a number.
Every access fact in business time and knowledge time. “What did we know, and when did we know it” for any date. Regulator-grade.
QuantVault is the action layer: orchestration, approvals, step-downs, hand-offs and execution back through your IAM and IGA. QuantCertify decides; QuantVault executes; nothing bypasses the record.
A read-only overlay on your existing identity and application stack. Nothing replaced, nothing re-architected. Builds on Entra, SailPoint, Saviynt and Okta.
04Derisking your technology
Technology risk in most firms is a narrative. In ours it is a number. We decompose every access and every agent into named risk factors across loss channels — cash, trading, regulatory, legal, reputational, IP — aggregate them cleanly from user to team to division to firm, and give you controls that act on the number before the loss.
Every access decomposed into named risk factors, priced in dollars, with covariance handled properly at every level of aggregation.
Dangerous entitlement composites detected semantically, including partial overlaps, cross-system combinations, and combinations that two agents would assemble between them.
Where rare, critical capability combinations live in one person; replacement coverage computed.
Cross-plane power concentration and shifts in performed role shape, explained in capability terms rather than raw event counts.
Threat signals mapped to business processes, producing surgical containment options instead of firm-wide shutoffs.
Onboarding an agent includes literally selecting its maximum daily damage. Halting grant issuance stops it instantly, with no revocation lag and no disruption to anything else.
05EU AI Act · Articles 12, 26 & 72
The EU AI Act’s high-risk regime is now binding law on fixed dates: 2 December 2027 for stand-alone high-risk systems — credit, insurance pricing, employment, critical infrastructure — and 2 August 2028 for AI embedded in regulated products. Penalties reach €15 million or 3% of worldwide turnover. Three articles form the chain: the system must record what it did (Article 12), the deployer must watch it, keep the record and be able to stop it (Article 26), and the provider must monitor it in the field, including how it interacts with other AI (Article 72).
The regulatory clock
In force; dates fixed.
Template due.
Annex III: credit, insurance, employment, infrastructure.
High-risk AI embedded in Annex I products.
As written, the Act asks for a record. Every earlier wave of automation moved from record-and-oversee to approve-before-acting once automated actions outran the humans watching them. QuantData was built for the second regime. The record the Act asks for is a by-product of the approval, which is why we meet the requirement and the ones that will follow it.
The checklist
Direct: met by a QuantData mechanism. Exceeds: goes beyond the text, usually by acting before the event. Shared: we supply the mechanism and evidence; you supply a judgment. Your process: an organizational act we make easy to evidence.
Showing 20 of 20 checklist items
| Requirement | Article | How QuantData handles it | Status |
|---|---|---|---|
| Automatic logging over the system’s lifetime | 12(1) | Level 0 observation ledger, append-only, cannot be configured off. Records intent, denials and non-events, not just what happened. | Direct — exceeds |
| Traceability appropriate to intended purpose | 12(2) | Every record bound to the business process step, object and declared purpose — traceability in business language. | Direct |
| Identify risk situations and substantial modification | 12(2)(a) | Drift decomposed into “environment moved” vs “agent moved”; manifest change-sets; fidelity violations; coverage degradation never silent. | Direct — exceeds |
| Logs feed post-market monitoring and deployer monitoring | 12(2)(b)–(c) | Observation API: query, subscribe, export. Per-agent, per-gate reporting with an evidence ID on every number. | Direct |
| Minimum log: period of use, reference data, matched input, who verified | 12(3) | Grant lifecycle in business and knowledge time; source route and version; hashed request; approver identity. The biometric minimum, met for every governed action. | Direct |
| Use the system per its instructions; unapproved systems cannot run | 26(1) | Manifest → human-approved operating surface → run mode. No surface, no grant, no action. The rogue agent is prevented, not discovered. | Direct — exceeds |
| Assign competent human oversight with authority | 26(2) | Accountable owner per agent; approval insertion routes questions to named approvers; override rates reportable. Competence and training remain yours. | Shared |
| Input data relevant to the intended purpose | 26(4) | Canonical concepts, official sources, version bridges, readiness checks, reconciliation. Semantic integrity enforced before the read. | Direct — exceeds |
| Input data sufficiently representative | 26(4) | We guarantee the system of record is what was read, at the right version and date. Accuracy of the values at source remains yours. | Shared |
| Monitor operation; suspend when use presents a risk | 26(5) | Aberrance ladder routes questions to people; halt-issuance and containment with zero lag. | Direct |
| Inform provider and market-surveillance authority; report serious incidents | 26(5) | Notification is yours. We supply the decision chain that establishes the causal link within hours. | Your process |
| Financial institutions: monitoring via internal-governance rules | 26(5) | Control rules are certified internal controls; the same evidence feeds the SOX AI Agent Control Reporting Pack. | Direct |
| Keep logs under your control for at least six months | 26(6) | Append-only, bitemporal ledger, retention by policy, held by you independent of any vendor’s logging. | Direct — exceeds |
| Inform workers; register systems; inform affected persons | 26(7)–(11) | Organizational acts. The agent registry and ledger give each one a precise basis. | Your process |
| Cooperate with authorities | 26(12) | Exportable, redaction-governed decision packets; every limit evaluation reproducible. | Direct |
| Post-market monitoring system, proportionate to risk | 72(1) | Control levels assigned per scope: observe everywhere, bound finance, grant on payment gates. The plan narrative is yours. | Shared |
| Collect and analyze performance for life; evaluate continuing compliance | 72(2) | Continuous ledger; drift and effectiveness analysis; readiness recomputed; non-compliance with the surface is a counted metric. | Direct |
| Analyze interaction with other AI systems | 72(2) | Interaction graphs mined; toxic combinations across agents blocked or routed to approval before the exchange completes. | Direct — exceeds |
| Human oversight: understand, interpret, override, stop | 14 | Every non-allow outcome is a question to a person; clarify-or-confirm forces a choice; halt-issuance is the stop button. | Direct — exceeds |
| Risk management: identify, evaluate, mitigate | 9 | Blast radius in dollars, chosen daily damage, toxic combinations; measures are pre-action. | Direct — exceeds |
We do not make your people competent, notify regulators for you, register your systems, test your model’s accuracy, or correct wrong values at source. We tell your people when they are needed and what the question is, and we produce the evidence for everything else.
Coverage by article
QuantEnt whitepaper assessment. Direct: met by mechanism. Shared: with your organization. Your process: evidenced by us.
06 · We’ve been down this road before
Knight Capital, 1 August 2012. A software change reactivated dormant code on one server. Millions of unintended orders. Close to the firm’s entire capital, before lunch. Knight had logs. Knight had people watching. It did not have a gate that could stop an order the firm never meant to send. The SEC’s pre-trade control rule already existed; Knight became its first enforcement case, and the industry never treated pre-trade controls as optional again.
SEC Rule 15c3-5: pre-trade risk controls become mandatory
Knight Capital: $440M lost in 45 minutes. Logs, yes. A gate, no.
First 15c3-5 enforcement action — against Knight. Pre-trade control is never optional again.
You would never let a trader trade without pre-trade controls. Why let ten thousand agents act without pre-action controls?
07Why QuantEnt
Everyone else records what happened. We decide what may happen, then record it. The difference is whether there is an incident.
Exposure, drift, blast radius and risk are measured and priced in dollars, not guessed. Built by the people who brought factor models to Wall Street.
An agent inherits the same authority model, the same limits and the same segregation of duties as the people it works beside. Two authority systems always disagree; we run one.
Read-only on your identity stack, above your data, through one gateway for agents. Nothing replaced. Nothing bypassed.
08Our products
Entitlements and business authority, compiled, priced and certified continuously. The land. Implemented and ready to deploy.
02The semantic brain and pre-action control plane for agentic AI. Four levels, from observation to per-operation grants.
03The action layer: approvals, step-downs, hand-offs and execution back through your IAM and IGA.
09Resources
Sand Hill East · Research note
A Sand Hill East research note on the EU AI Act’s Article 12, where the regulation is going, and what industries should do about it.
QuantEnt · Whitepaper
The QuantEnt whitepaper: Articles 12, 26 and 72 — and 9, 14, 19 and 73 — clause by clause, with the coverage and the boundaries stated plainly.
Industry outlook
Inadequate risk controls are a leading cause.
Source: Gartner, June 2025. Forecast.
10About QuantEnt
QuantEnt was founded to solve a problem we saw repeatedly inside large, regulated organizations: nobody could say what anyone was actually allowed to do, and nothing checked before the action. We built the answer for people first — entitlements compiled into one computable model, priced in dollars — and then for agents, because an AI that cannot be bound to a mandate cannot be trusted with one. Our CEO ran Barra, the financial industry’s most successful factor platform. He is now bringing factors to entitlements, and pre-action control to AI.
Founder & CEO
Trent Walker is a senior technology executive and former hedge fund CTO with nearly three decades of experience building and leading mission-critical platforms for trading, risk, finance, and control functions across buy-side, sell-side, and market infrastructure. Most recently at Point72, he served as Head of Risk & Controllers Technology, modernizing core systems, strengthening data and process coherence, ensuring robust P&L foundations, and advancing frameworks for equity factor analytics and risk calculation.
Previously, Trent was Head of Technology Strategy at Nasdaq and a Managing Director at MSCI, where he led application development for the Barra and RiskMetrics platforms. He also served as CTO at BlueCrest Capital Management and held senior technology leadership roles at Credit Suisse, helping drive the shift from spreadsheet-driven workflows to scalable front-to-back systems and disciplined new-product onboarding. He ran Barra, the financial industry’s most successful factor platform.
Trent has also led in the front office. At Barclays Capital he was Global Head of Risk and Margin for Prime Services, leading global netting and risk teams overseeing over $1 trillion in financing products and building a coherent, legally accurate view of counterparty exposure and set-off. This blend of front-office accountability, deep engineering experience, and still being a hands on developer gives Trent a rare perspective on how desk-level decisions propagate into data, models, controls, and firmwide outcomes.
Trent began his career as an Assistant Professor of Mathematics at UC Santa Barbara after earning his PhD from UC Berkeley, specializing in operator algebras and control theory. In his spare time, Trent invented the patented Induction Press coffee machine, launching in Q2 2026.
CTO
Justo Ruiz is an AI founder and hands-on computer scientist specializing in distributed systems, high-performance computing, and generative-AI infrastructure, with a track record of shipping production platforms for data-intensive workloads across startups and top hedge funds.
As Founder and CEO of Shapelets, Justo built a high-performance data processing and visualization platform and architected a vector database capable of ingesting 100,000+ embeddings per second for real-time semantic search and generative-AI use cases.
Previously, as CTO of a private equity–backed company, Justo reset product strategy, established development centers in Spain, modernized the platform, and contributed to a successful exit.
In front-office technology roles, including as Principal at BlueCrest Capital Management, Justo unified market and reference data systems and delivered a consolidated high-performance risk platform for trading and portfolio decisions.
Earlier at Barclays Capital as a Senior Quant Developer on the Convertible Bonds desk, Justo built performance-critical pricing and trading infrastructure for complex hybrid securities.
11Talk with us
Most engagements begin one of three ways: a priced scorecard of your entitlements, an observation-only deployment of QuantData across your agent estate, or a control-level plan against the December 2027 date. Tell us which, and we will route you to the right people.