Danish software company cBrain has introduced F2 AI Case Preparation, a technology designed to run AI across large government information systems while cutting computing requirements and keeping data within an authority's own controlled environment. The company said the approach was successfully demonstrated earlier this year in a proof-of-concept project with the State of California, working with environmental cases containing documents of hundreds of pages.
The technology takes a different route from the current AI research focus. Instead of building larger language models that can process bigger context windows, F2 AI Case Preparation reduces and prepares the context before it reaches the AI model. It uses semantic search and vector-based information retrieval to identify, retrieve, structure and rank the information relevant and authorized for a specific administrative task.
"Rather than extracting and bringing large volumes of government data to the language model to determine what matters, F2 brings the AI model to the government data and prepares what matters before invoking the model," the company said in its announcement.
Why context size alone doesn't solve government AI
A single government case can contain hundreds of pages accumulated across documents, correspondence, decisions, hearings and regulatory material. An entire authority's information domain may consist of millions of documents. Simply expanding what a language model can process doesn't address the fundamental challenge of applying AI efficiently across those very large domains.
If an AI model is given thousands of pages just to determine which pages are relevant to a particular task, significant computing resources get spent processing information that contributes little or nothing to the answer. F2 AI Case Preparation moves that selection process into the government platform itself, reducing the amount of information the language model needs to process.
The technology builds on cBrain's existing F2 cSearch, which uses Apache/Lucene technologies to index large volumes of data. Because F2 is the platform where cases, documents, processes, metadata, users and access rights are already managed, the selection process can happen inside the governed environment rather than in an external AI tool.
Data sovereignty and the California project
Government organizations handle information that may be confidential, sensitive or subject to regulatory requirements. F2 AI can operate on-premise, keeping government information within the authority's controlled environment. Cases, documents, correspondence and metadata, as well as processes, organizational structures and access rights, all remain governed by existing authorization mechanisms.
In February 2026, cBrain announced an informational proof-of-concept project with the State of California focused on applying AI and Natural Language Processing to large volumes of environmental and regulatory information. The project has now been completed successfully, with F2 AI Case Preparation applied across complex government cases containing documents with hundreds of pages of CEQA environmental data.
In parallel, the F2 platform has been added to the State of California's Software Licensing Program (SLP), a statewide procurement framework designed to simplify software purchases by California government entities. cBrain sees the SLP inclusion as an important step in supporting its further market development in California.
Why this matters for government professionals
For government agencies considering AI deployment, the practical takeaway is that large language models don't have to read everything to be useful. The California demonstration shows a working alternative: prepare the relevant context first, then invoke the model on a smaller, focused set of information. That approach can reduce hardware costs and keep sensitive data inside the agency's own infrastructure, which matters when dealing with confidential or regulated information. For AI for Government professionals weighing implementation options, this points to a procurement consideration: the platform you already use for case management may determine how efficiently AI can be applied to your data. Policy makers evaluating AI proposals may find it useful to ask vendors how their technology handles context selection and data sovereignty before committing to large-scale deployments - questions that an AI Learning Path for Policy Makers can help frame.
Your membership also unlocks:
