From Risk Data Silos to Autonomous AI: Future‑Proofing Bank Risk Architecture for 2026 and Beyond

As banks move from one‑off regulatory implementation projects into a steady‑state world of continuous change, they face an unprecedented surge in risk data volumes, complexity, and expectations. The Fundamental Review of the Trading Book (FRTB) has been a particular catalyst in the Traded Risk world: ongoing uncertainties around timelines and national discretions have forced banks to rethink not just their new models, but existing ones as well - especially for market risk VaR - including the underlying data and architecture that support them.

At the same time, supervisors have not relaxed their expectations around risk data aggregation and reporting. BCBS 239 and the broader Risk Data Aggregation and Risk Reporting (RDARR) agenda in Europe remain a benchmark for how banks should organise, control, and exploit their risk data across entities, risk types, and time horizons. Far from being “done”, RDARR is being reinforced in recent supervisory guidance and assessments, underlining that the ability to aggregate and explain risk data quickly and consistently is now a hard requirement rather than an aspiration.

Layered onto this are broader regulatory themes: the finalisation of Basel III (Basel 3.1), climate and ESG risk requirements, and evolving liquidity and funding rules. All of these converge to put risk data infrastructure under sustained pressure.

Technology, however, has caught up. Cloud‑scale architectures, modern columnar storage, real‑time analytics, and AI‑driven insight now make it possible not just to cope with this explosion in data, but to turn it into a strategic advantage. The core question going into 2026 is no longer simply “how do we store all this data?” but “how do we organise, govern and expose it so that every risk, finance and trading user - human or AI-assisted - can interrogate all of it, on demand, at sustainable cost and with full explainability?”

What follows outlines how banks can move from fragmented, regulation‑specific solutions towards a unified, AI‑powered risk data platform.

From one‑off FRTB projects to a strategic, AI-powered risk data platform

Calculating market and credit risk accurately has always been fundamental for banks. What has changed is the depth, frequency, and traceability that regulators and management now expect. The revised market risk framework alone has driven a step change in the granularity and length of histories required for sensitivities, risk‑theoretical P&L, modellability assessments, non‑modellable risk factors, and stress scenarios.

Crucially, these requirements now sit alongside other data‑intensive regimes: enterprise‑wide stress testing, climate and ESG risk, granular credit and liquidity metrics, and more detailed prudential and statistical reporting. Each new theme tends to arrive with its own data model, timelines, and quality standards.

Most banks have already felt the consequences:

• Ballooning data warehouses and lakes
• Multiple overlapping feeds and reconciliations
• Proliferation of point solutions built to meet a single regulation or deadline
• Growing infrastructure costs and operational risk
• Slower risk calculations and reporting cycles than the business would like

Going into 2026, the challenge is to consolidate what was built for specific projects like FRTB, stress testing, and regulatory reporting into a coherent, future‑proof risk data platform – one that is ready not just for the next regulation, but also for secure, governed AI use cases.

Without this foundation, AI in risk tends to remain a collection of prototypes, limited to narrow datasets and disconnected from production decisionmaking.

Capital calculation: complexity, granularity and the data burden

Moving back to the revised market risk framework, banks still choose between – and often run in parallel – the standardised approach and the internal models approach (IMA). But the data implications of that choice have grown dramatically.

For IMA in particular, requirements around:

• Risk‑theoretical and hypothetical P&L
• Modellability tests and non‑modellable risk factor capital
• Desk‑level model approval and ongoing backtesting

all drive a substantial increase in both transactional and historical data. Even banks that rely predominantly on the standardised approach face higher data needs to support richer risk factor histories, scenario analysis, attribution, and governance.

For cross‑jurisdictional groups, this is compounded by slightly different local implementations of global rules. Global banks must manage:

• Multiple jurisdictional rule sets and reporting templates
• Proxy data and reference data inconsistencies
• Booking model variations across entities
• Full auditability and versioning of both data and methodologies

What emerges is not just “more data”, but more interconnected, interdependent data pipelines that are difficult to manage with legacy, silo‑based architectures.

From fragmented datasets to a unified, explorable, and AI-powered risk data foundation

Historically, technology limitations pushed banks towards fragmentation. Separate infrastructures were the norm for:

• Market risk, credit risk and liquidity risk
• Daily risk and long‑term historical archives
• “Normal” and stressed datasets
• Production reporting and ad‑hoc analysis environments

In‑memory technologies and traditional warehouse designs enforced hard constraints: if you wanted performance, you had to limit history; if you wanted history, you had to move data into separate, slower stores; if you needed a new view, you copied and reshaped data into yet another silo.

Modern horizontally scalable architectures remove many of these constraints. With the right data model, banks can now aim for a single logical risk data platform that supports:

• Market, credit and liquidity risk on the same underlying data
• Daily, intraday and multi‑year histories without artificial splits
• Normal and stressed datasets in one place, rather than duplicated copies
• Regulatory, management, and trading analytics off a common golden source

This shift is not just technical – it transforms what users can do. In a unified architecture, a risk manager can, in a single environment:

• See a country’s contribution to VaR and expected shortfall
• View exposure at default and other credit metrics for the same country
• Overlay stress losses, sensitivities, and liquidity measures
• Link all of this to P&L information and drill down to trade level
• Roll up to portfolios, entities, and group views at will

At the same time, an agentic AI solution can be safely pointed at this governed, well‑organised risk data foundation to:

• Answer simple or complex questions combining potentially multiple risk types
• Leverage various risk focused agents across very large datasets to offer valuable insights to users
• Suggest relevant analysis and reports

All of this should be possible without waiting days for bespoke extracts or manually stitching spreadsheets from different systems. When relevant data sits in a unified, queryable structure with clear lineage and controls, both humans and AI tools can rely on it – reducing the need for manual reconciliations and joint reports assembled, lowering operational risks.

Data storage optimisation: consolidation without compromise

In practice, many banks are already on this journey. For example, Opensee has worked with a large international bank to redesign its risk data architecture around these principles offering its 500+ users a modern and robust platform.

The starting point was a landscape of multiple very large datasets – covering market risk, credit risk, historical archives, and stress copies – totalling several hundred terabytes. Each dataset had been built for a specific purpose, often under tight regulatory deadlines, resulting in substantial overlap and duplication.

By consolidating these into a single logical data model with real‑time access, the bank was able to:

• Combine multiple major datasets into one coherent structure
• Remove a significant proportion of duplicated data points at ingestion, immediately reducing storage and infrastructure costs
• Streamline adjustments and reconciliation processes by eliminating redundant copies of the same information
• Give users direct access to both market and credit risk views from the same platform, regardless of the required granularity or history

An abstraction or semantic layer shielded users from the complexity of the underlying schema. Instead of understanding every table, key and join, risk and business users could focus on the questions they needed to answer: capital, sensitivities, concentrations, P&L explain, liquidity metrics, climate scenarios, and more. The system handled the technical work of pulling from the correct underlying data.

For AI interaction, this abstraction layer is equally powerful. It provides a controlled way for AI models or assistants to query data via governed APIs, semantic layers, or vectorised representations, preserving performance and security, simply transforming natural language questions into API calls.

This approach shows how a single, scalable platform that optimises daily storage and compute can transform the risk data lifecycle. Longer, more consistent histories become feasible, enabling:

• More meaningful trend analyses
• Better alignment between stress testing and daily risk management.
• Stronger data lineage and explainability for both internal and external stakeholders

At the same time, AI-based data quality improves as reconciliations move closer to source and fewer copies exist to go out of sync, a prerequisite for any serious use of AI in risk management.

Real‑time, self‑service analytics as a differentiator

Looking ahead to 2026 and beyond, expectations will only increase. Supervisors are pushing for faster, more frequent and more granular views of risk and capital. Boards and senior management expect forward‑looking analytics, scenario capabilities, and the ability to interrogate exposures in near real time. Front‑office and business units want on‑demand insight into capital consumption, risk‑adjusted profitability, and constraints.

Meeting these expectations with batch‑oriented, siloed architectures is not sustainable. A future‑proof approach to risk data architecture must deliver:

• Real‑time or near real‑time access to complete risk datasets, including intraday where needed
• Self‑service analytics, so risk, finance and front‑office users can slice, dice, and aggregate at any level without relying on specialist teams or offline extracts
• Horizontal scalability, to absorb new regulatory requirements, more scenarios, longer histories and new business lines without wholesale redesign
• Cost‑efficient storage and compute that actively eliminate duplication and expensive copies while preserving performance.

Once this is in place, AI can be used not as a separate “innovation silo”, but as a natural extension of the platform. Examples include:

• AI-driven anomaly detection on sensitivities, exposures, or P&L explains across billions of rows
• AI assistants that allow risk managers to query complex datasets in natural language and receive answers tied to traceable data queries
• Scenario generation and enrichment, where AI helps design plausible but severe scenarios that are then calibrated and validated using the underlying structured data
• Intelligent reporting and analysis gathering, where AI agents automatically compile the data extracts and explanations needed for an efficient monitoring and regulatory reporting

When banks tackle exponential data growth with this kind of unified, scalable, and AI-ready architecture, they open the door to rethinking the entire risk data structure. Instead of building a new silo for each new regulation or AI experiment, they can plug additional use cases into a common platform – from FRTB and IRRBB to climate scenarios, ESG metrics, funds transfer pricing, and beyond.

Turning regulatory pressure and AI momentum into a strategic advantage

Regulatory change has often been seen as a cost of doing business – necessary, but not strategic. The wave of FRTB implementation, Basel III finalisation, climate risk expectations and digital resilience rules could easily be viewed the same way. The same is true of AI: widely discussed, often presented either as a risk in itself, or as a collection of proofs of concept disconnected from core processes.

However, for banks willing to use this moment to modernise their risk data architecture, the story can be different. By consolidating fragmented datasets, embracing horizontally scalable technologies and empowering users with real‑time, self‑service, and AI-assisted analytics, institutions can:

• Reduce infrastructure and operational costs
• Improve data quality, consistency, and explainability
• Accelerate risk and capital calculations and reporting
• Enable richer, more forward‑looking risk insights for decision‑makers
• Deploy Agentic AI solutions on top of governed, high-quality data rather than ad hoc extracts
• Create a single risk data foundation that can adapt as regulations, business models, and AI capabilities evolve.

And finally realise the intent of RDARR in day‑to‑day practice: timely, accurate, comprehensive risk aggregation, and truly consistent reporting.

Looking forward, this RDARR aligned foundation is what enables the next step change: autonomous AI assistants operating directly on governed risk data to create actionable insights, narratives, and reports. Instead of risk teams manually stitching together extracts, spreadsheets, and slide decks, banks will increasingly rely on AI agents that can:

• Pull the right data from the unified platform
• Run the relevant calculations and scenarios
• Detect anomalies or emerging risks
• Assemble clear explanations and visualisations tailored to different audiences – from trading desks to the board and supervisors

In this model, risk professionals spend less time on data wrangling and report production, and more time on judgement, challenge, and decision‑making. Supervisors see faster, more transparent responses. Senior management gains near real‑time visibility on capital, liquidity and emerging risks, supported by consistent, explainable analytics.

In that sense, the uncertainty and pressure around FRTB, RDARR, and other regulations combined with the rapid evolution of AI technologies may prove to be a catalyst rather than just a burden: a unique opportunity to move from fragmented, compliance‑driven infrastructure to unified, analytics and AI-ready risk data foundations that support both regulatory obligations and strategic decision‑making in 2026 and well beyond.