S.AI.F
Book a demo
About S.AI.F · Secure AI Foundation

We exist so that trust in AI can be proven, not asked for.

S.AI.F is an independent validation, security and policy house headquartered in Abu Dhabi. We test AI systems against declared thresholds, protect the compute they run on, and issue signed proofs that anyone can verify — for banks, ministries, hospitals and the labs that build the models.

Boardroom with a glowing constellation model and the Abu Dhabi skyline
Mission
Make every consequential AI deployment provably safe, secure and compliant.
Method
Independent testing against thresholds declared in advance. Confidential compute so we never see plaintext weights. One signed Trust Proof binding model, data, compute and results — verifiable without contacting us.
Mandate
Serve the UAE's national AI programme as its reference validator, then extend the same standard to the Gulf and to any jurisdiction that needs conformity it can check rather than believe.
Why we started

The world was spending on capability a hundred times faster than on proof.

By 2025 the region was deploying AI in credit, health, public services and infrastructure at a pace no other market matched. What did not exist was an institution that could say, with evidence a regulator would accept, that a given system was fit for that use — and stand behind the statement with a signature.

We had done pieces of this work inside central banks, national labs and audit firms. Each time the same gap appeared: excellent builders, well-meaning policies, and no independent chain of evidence between them. S.AI.F was founded to be that chain.

We chose Abu Dhabi because the ambition, the data sovereignty requirements and the institutional speed were all here. We chose independence — structurally, not rhetorically — because a proof from the people who sold you the model is not a proof.

The independence line

Four commitments that make our signature worth something.

01
We never build what we assess
No model development, no integration work, no reseller agreements with vendors we test.
02
Validators are paid by the firm, not the file
No commission on engagements. A failed gate costs a validator nothing; a wrongly passed one costs their signing key.
03
Thresholds are declared before results
Every pass condition is fixed in the assessment plan and hashed into the proof before a single test runs.
04
Anyone can check our work
Proofs verify with an open tool and a public key. Our methodology and control mappings are published.
Trusted by

Institutions in the UAE that have relied on our outcomes.

ILLUSTRATIVE · REPLACE WITH LICENSED LOGOS AND SIGNED REFERENCES
mubadala
Sovereign investment
Portfolio-wide AI governance baseline: model register across eleven holdings and an approval template every board now uses.
Policy42001 mapping
ADNOC
Energy
Predictive-maintenance and safety-critical vision models validated for robustness and drift; monitoring bound to approval thresholds.
ValidationTrust Proof
Emirates NBD
Banking
Credit-decision models tested per protected group against declared fairness thresholds; recourse route redesigned and measured.
FairnessTrust Proof
e&
Telecoms
Customer-service LLM red-teamed against OWASP LLM Top 10; containment controls verified before national rollout.
Red teamSecurity
mashreq
Banking
Onboarding and fraud models assessed inside confidential compute; post-quantum signed proofs shared with the regulator.
Secure computeTrust Proof
Al-Futtaim
Retail & automotive
Pricing and recommendation systems reviewed for explainability and consumer-protection alignment across four markets.
ExplainabilityPolicy
6
Reference institutions across energy, finance, telecoms, retail and sovereign investment
41
Models assessed to a signed Trust Proof since first issue
0
Proofs withdrawn for a finding we should have caught
100%
Assessments run inside confidential compute in the UAE
Where we are going

Prove, issue, accredit, set the standard.

  1. 2025 · Q4
    Founded in Abu Dhabi
    Core team of seven from central banking, national labs and audit. Methodology and proof format published.
  2. 2026
    Prove
    First Trust Proofs issued for finance and energy. Confidential-compute assessment cell live in the UAE.
  3. 2027
    Issue at scale
    Thirty-person practice. Public register of certified systems. ISO/IEC 42001 certification of our own management system.
  4. 2028
    Accredit and set the standard
    Accredited private validators. Gulf-wide recognition of the proof format. Reference input to national and regional standards.
How we work
Evidence over assurance
We do not tell clients their systems are fine. We show what was tested, what was found, and what changed.
Sovereignty by design
Weights, data and results stay in the UAE, inside confidential compute, under keys the client controls.
Humans stay accountable
Every proof carries a named signature. Every automated decision we certify has a route to a person.
Publish the method
Control mappings, test suites and the proof format are open, so trust in us is never a prerequisite for trusting a proof.
From the journal

Recent essays

All insights →
AI governance

AI governance is not a policy document. It is a chain of evidence.

Responsible AI

Responsible AI in practice: five tests that separate intention from outcome

Future of AI

The future of AI in three horizons — and what each one asks of us

Work with us, or join us.

Clients start with a 30-minute scoping call. Candidates start with a work sample. Both end with a person who will put their name to the outcome.

Book a demoOpen roles →