KERVALT / ETHICS
Responsible AI by design
Kervalt's ethics program sets the principles, review boards, and checks that guide how our systems are built and used.
01 / PRINCIPLES
Ethics principles
The standards that guide every research decision and product release.
Accountability
Every answer can be traced back to who asked, what version was used, and where the information came from. We do not build systems that hide their origin.
Fairness
We test our systems for unfair differences across groups and places. We publish what we find and how we fix it.
Transparency
Model cards, test reports, and safety results are published so customers and regulators can check our claims.
Privacy
Your data is never used to train public models. All processing happens on secure European servers with strict access controls and logs.
Human oversight
Important decisions are designed for human review, with clear escalation paths and override controls.
Societal impact
We review possible harms before releasing new capabilities and decline projects that violate our charter.
02 / PRACTICE
From principles to practice
Kervalt runs an internal ethics review board, publishes audit results, builds fairness checks into development, and requires impact assessments for new capabilities.
BEFORE
Ethics as marketing language
Many vendors publish vague statements without review boards, audits, or enforcement. Fairness tests are optional. Customer data feeds model improvement. Impact assessments never happen.
AFTER
Ethics as an operational system
Kervalt runs an internal ethics review board, publishes audit results, builds fairness checks into development, and requires impact assessments for new capabilities. Customer data never trains public models.
03 / BIAS MITIGATION
Fairness checks for each part of the system
Each part of Kervalt is tested for unfair bias.
Tafari
Fairness in search results
What it does: We check whether search results favor or ignore certain groups, places, or document sources.
Why it helps: Catch skewed evidence before it shapes a decision.
What you get: Decisions are grounded in balanced, representative information.
Griot
Fairness in connections
What it does: Analysis detects under-represented people or organizations and uneven relationship coverage.
Why it helps: Prevents blind spots in ownership, supply-chain, or influence networks.
What you get: Risk models reflect the real world more accurately.
Elimu
Fairness in numbers
What it does: Tests outcomes across protected attributes for unfair differences.
Why it helps: Measures unfair impact with clear statistical tests.
What you get: Compliance and HR analytics meet fairness standards with auditable proof.
04 / STEWARDSHIP
Your data is yours
Customer prompts and documents are never reused to improve third-party models. Your data remains under your control.
PICTURE
Data misuse
Some services reuse customer documents and questions to improve outside models, creating conflicts with professional secrecy.
PROMISE
Data dignity
Your data remains under your control. We do not train public models on customer workloads.
PROVE
Contractual guarantees
Data-use terms, EU hosting, and ISO 42001 alignment are written into agreements, not footnotes.
PUSH
Review our charter
Download the full ethics charter and contact our ethics team with questions.
NEXT STEP
Hold us accountable
Read the ethics charter, request an audit summary, or raise a concern with our ethics team.