Squirro updates AI platform for regulated enterprises

Squirro has released its May 2026 Long-Term Support platform update, aimed at organisations using AI in regulated settings.

The update focuses on three areas the company identifies as barriers to wider corporate AI adoption: unreliable answers, compliance risk and the difficulty of deploying systems across large organisations. The new version is designed for production use in institutions where data access controls and auditability are critical.

Among the changes is a system intended to keep AI responses tied to defined contextual data such as a client, project or dataset. That approach is meant to reduce the risk of out-of-context answers when tools are used across large workforces and more complex workflows.

Dave Clarke, Chief Executive Officer of Squirro, linked the release to growing pressure on companies to show commercial returns from AI spending. "The enterprise AI conversation has shifted from broad capabilities to measurable bottom-line impact," he said.

He added: "This LTS release delivers the granular precision and deployment agility required to help organisations expand their AI capabilities across the enterprise, and accelerate core business cycles while significantly reducing the total cost of ownership for global AI infrastructure."

Accuracy focus

The platform now uses entity-based filtering and revised agent tool execution to keep responses grounded in the correct source material. Squirro said this is intended to improve output in multi-step tasks that need to draw on specific records or repositories rather than broad sets of enterprise data.

That matters most in sectors where incorrect answers can create operational or regulatory problems. Squirro lists customers including the European Central Bank, the Bank of England, Standard Chartered Bank, Oversea-Chinese Banking Corporation and Henkel, all of which operate in environments where internal controls and data traceability carry particular weight.

Jan Ebner, Head of Product at Squirro, said the issue becomes more acute as deployments expand. "As our clients scale complex use cases across thousands of users, the margin for error shrinks to zero," he said.

He continued: "By introducing entity-based filtering and refining our agentic tool execution, they can use the AI with the right tools for complex, multi-step reasoning with unprecedented stability, ensuring answers are grounded in and validated against the right data."

Governance controls

A second part of the release centres on zero-trust governance. Squirro has added all-permissions-match validation logic intended to prevent unauthorised data exposure across multiple silos and hierarchies inside large organisations.

The update also integrates existing access control lists directly into the platform so governance rules apply as data enters the system. In practice, that means permissions already used by an organisation can carry through into AI search and workflow tools rather than being recreated in a separate layer.

Such controls have become more important as companies move from pilot projects to broader internal rollouts. In regulated industries, a wrong answer is not the only concern. Systems must also show that users only see information they are authorised to access and that those restrictions remain consistent across departments and jurisdictions.

Deployment changes

The release also includes changes aimed at IT and DevOps teams. Squirro now supports single-node containerisation for more standardised deployments, a step intended to fit more easily into continuous integration and continuous deployment workflows.

According to the company, that should reduce the time needed for testing, validation and upgrades. The platform also now includes full Unicode compatibility through utf8mb4 and uses Apache Tika 3.2.3 to parse international characters and complex file types.

Those lower-level changes matter because enterprise AI systems often depend on pulling information from a wide range of internal documents, archives and repositories. Problems parsing files or handling multilingual text can undermine answer quality before any model produces an output, particularly in global organisations managing records across different systems and languages.

Founded in 2012, Squirro has teams in Switzerland, the United States, the UK and Singapore. The company positions its software around enterprise search, workflow automation and AI tools for regulated industries, with a focus on auditable outputs and permission-aware access to information.