Company
Building the audit layer for accountable AI agents
Arbiris exists because regulated firms are beginning to deploy AI agents into workflows where ordinary logs are not enough. The company turns agent activity into evidence a Senior Manager, compliance team, and regulator can understand.
Current position
- MVP status
- MVP 2 built
- Signed records and policy binding shipped; evidence-pack methodology independently reviewed and refined.
- Stage
- Design partners
- Pre-revenue and focused on first regulated financial services partners.
- Review
- Ex-FCA/FSA specialist
- Evidence-pack methodology reviewed by an ex-FCA/FSA specialist, with findings incorporated.
From policy to runtime evidence
Existing tools leave a gap. Engineering observability platforms are useful for debugging, but not built to produce evidence a regulator can readily assess
FCA SM&CR makes named Senior Managers accountable for decisions made inside their firms. As agents begin to plan, call tools, make recommendations, and affect customer outcomes, firms need evidence that can survive compliance review: not just what the model said, but what the agent intended, which policy applied, what data it used, and who remained responsible.
GRC platforms help manage policy, but they rarely capture agent-level intent, delegated action, or human oversight at runtime. Arbiris sits in that gap. The product combines AARF, an open accountability record standard, with developer instrumentation, signed intent records, a compliance-facing dashboard, and evidence packs for review.
Founded by an AI and data engineer
Darren Gidado
Founder
Arbiris is built by a founder who cares about data — a data science specialist and the architect behind a transaction-scoring system for an FCA-regulated Open Banking fintech
His experience also spans data pipelines used for commercial decision-making. The founder case is technical and execution-led: Darren built the Arbiris MVP pipeline around signed audit records, policy binding, and agent intent capture. Two Arbiris MVPs have shipped as a solo founder, with AARF published as an open standard and MVP 2's evidence-pack methodology independently reviewed by an ex-FCA/FSA specialist.
Operating principles
Evidence should be legible to compliance teams, not just engineering teams.
Agent accountability records should be tamper-evident, portable, and framework-agnostic.
Regulated firms should be able to explain AI agent actions and accountability.
What's verifiable today
- Published standard
- AARF is public
- The accountability record format is released under CC BY 4.0, so design partners can review, adapt, and challenge the structure rather than taking a black-box framework on trust.
- Regulatory mapping
- Built around named accountability
- The standard is mapped to FCA SM&CR, the EU AI Act, UK GDPR, and ISO/IEC 42001, with the focus on who authorised, monitored, and remained accountable for agent behaviour.
- FCA programmes
- Academy offer + 2 applications
- Offered a place in the FCA AI Lab's Supercharged Academy; applications submitted to the Digital Sandbox and AI Spotlight programmes. None of this should be read as FCA endorsement of Arbiris or its product.
Pre-revenue, design partner stage
Arbiris is pre-revenue and actively looking for its first design partners — organisations who take this category as seriously as it does