LuSarv Labs · Research
The accountable alternative to AI as usual.
LuSarv Labs is a research programme with a single focus: making AI accountable in the settings where it matters most. We work on system designs in which outputs are grounded in trusted evidence, checked for important omissions, and supported by an auditable record before they are relied upon.
The problem
The liability gap
When an AI-assisted decision causes harm, who answers for it? Today's systems offer fluency and confidence scores, but no verifiable record of why an output should have been trusted. That gap, between what AI produces and what accountability demands, is where LuSarv works.
Better prompting, generic guardrails or a stronger base model can support reliability claims. Our position is different: high-stakes trust depends on how the overall system is designed for accountability, sovereignty and scrutiny. Accountability is achieved by system design, not by policy alone.
Research direction
What accountable AI requires
Four commitments define the programme. We publish the position publicly; the mechanisms themselves remain internal while the company operates in stealth mode.
01
Hypothesis generators, not oracles
Large language models produce fluent candidates, not verified answers. LuSarv's research treats every LLM output as an assumption that must earn reliance; the model proposes, the evidence decides.
02
Grounding in trusted evidence
Outputs in high-stakes settings must be anchored in source material the organisation already trusts. Our direction of research is systems where no claim reaches a user without verifiable grounding.
03
Checked for material omissions
A correct partial answer can be as dangerous as a wrong one. We investigate how systems can be held to account not only for what they assert, but for what they fail to surface.
04
An auditable record before reliance
Trust in high-stakes environments depends on a timestamped, tamper-evident record of what was generated, what it was checked against, and what was relied upon, forming a legally defensible audit trail by design.
Evidence on hand
The white paper
The current limitations of AI in high-stakes decision environments
The public articulation of the liability-gap thesis, and the entry point to the broader research vision. It is written for the decision-makers who must defend AI adoption to governance, procurement, clinical safety and audit, and it is shared on request rather than as an open download.
Collaboration
Working with NHS and academic partners
We welcome research conversations with NHS trusts, academic health systems and grant-linked programmes where accountability is a requirement, not an afterthought.