RESPONSIBLE AI

Responsible AI

How we build and use AI at QuantumLayer

01APPROACH

Our approach

QuantumVerifi uses AI systems to assist software engineers in analysing code, generating tests, identifying security vulnerabilities, and improving software quality. Our AI is a tool that augments human expertise — it does not replace engineering judgement.

02TRANSPARENCY

Transparency

—All AI-generated outputs are clearly identified as machine-generated
—Analysis results include provenance metadata — which model, which version, when it ran
—Security findings are labelled with confidence levels and severity ratings
—Compliance evidence chains provide cryptographic proof of exactly what was executed
03OVERSIGHT

Human oversight

AI-generated results — including tests, security findings, documentation, and performance reports — are probabilistic outputs. They may contain errors, false positives, or omissions.

—All AI outputs should be reviewed by a qualified human before use in production
—Security findings are informational — they do not constitute a professional penetration test
—Generated tests validate behaviour at the time of analysis and may need updates as code evolves
—We recommend using QuantumVerifi as part of a broader quality assurance strategy, not as a sole gate
04DATA PROTECTION

Data protection

—We do not use customer code or data to train general-purpose AI models
—Per-tenant model adapters (available on Scale/Enterprise plans) are isolated — your data is never used for other customers
—Source code is processed temporarily during analysis and is not stored permanently
—LLM observability data is hashed by default — full prompt text is opt-in only
05VALIDATION

Safety and validation

Every AI-generated artefact goes through a multi-stage validation pipeline before being presented to users.

—Syntax validation via tree-sitter AST parsing (165+ languages)
—Compilation checking before execution
—Sandbox execution in isolated environments with resource limits
—Self-healing pipeline automatically fixes common errors (up to 3 attempts)
—Tests that fail validation are flagged, not silently included
06IMPROVEMENT

Continuous improvement

We continuously evaluate and improve our AI systems. This includes monitoring output quality, tracking validation pass rates, and updating prompts and models as better options become available. We use structured observability (via Langfuse) to measure quality metrics without exposing customer data.

07LIMITATIONS

Limitations

QuantumLayer Platform Ltd does not guarantee the accuracy, completeness, or fitness for purpose of any AI-generated output. AI systems can produce incorrect results, miss edge cases, or generate code that appears correct but contains subtle bugs. Use of AI-generated results is at your own discretion and risk.

08CONTACT

Contact

Questions about our AI practices? Contact us at: [email protected]

← Back to homeTrust Centre →