Products.
The public repositories address different parts of the verification problem, from stating properties to checking and mediating execution.
The original project catalogue, technical topics, and source links are retained. Its status classifications and feature descriptions describe the earlier site, not an independently verified release or guarantee.
Platform
- 1.1
Provability Fabric
Research on agent execution controls, policy decisions, proof-related artifacts, and records suitable for later examination.
- Runtime verification
- Proof-carrying code
- Policy engine
- Rust · WASM
Original project description and classification
The foundational runtime for proof-carrying behaviour. Intercepts agent actions, validates attached proofs against policy, and admits only verified operations.
Original catalogue label: Core. This label is retained for continuity and is not a current certification.
Build & verify
- 2.1
Spec-to-Proof
Tools for examining the translation of natural-language specifications into formal obligations and Lean artifacts.
- LLM-to-Lean
- Policy compilation
Original project description and classification
Compile natural-language specifications into Lean theorem proofs. Transforms policy documents into machine-checkable mathematical artefacts.
Original catalogue label: Active. This label is retained for continuity and is not a current certification.
- 2.2
SpecCursor
Dependency-upgrade workflows that examine change obligations, verification results, and their relation to existing specifications.
- Dependency safety
- Change verification
Original project description and classification
Formal qualification for dependency upgrades. Ensures code changes preserve existing behavioural guarantees through automated proof verification.
Original catalogue label: Active. This label is retained for continuity and is not a current certification.
- 2.3
SpecSync
Specification coverage and verification checks associated with software changes and pull requests.
- GitHub Actions
- CI/CD gates
Original project description and classification
GitHub-native specification coverage and verification gates, integrated into pull-request workflows.
Original catalogue label: Active. This label is retained for continuity and is not a current certification.
- 2.4
Lean Toolchain
Lean 4 libraries and utilities used to develop and inspect formally stated properties.
- Lean 4
- Verification library
Original project description and classification
Curated Lean theorem prover distribution with extensions for AI safety verification workflows.
Original catalogue label: Active. This label is retained for continuity and is not a current certification.
Runtime & evidence
- 3.1
Runtime Safety Kernels
Formal specifications and implementation work concerning the trusted components used to mediate AI execution.
- Isolation
- Hardware security
Original project description and classification
Minimal trusted computing kernels provide hardware-level isolation for AI workloads with a reduced attack surface.
Original catalogue label: Active. This label is retained for continuity and is not a current certification.
- 3.2
Security Envelopes
Deployment-boundary properties, including access control, isolation, and attestation interfaces.
- Encrypted execution
- Confidential computing
Original project description and classification
Encrypted execution environments providing confidentiality and integrity guarantees for AI model inference and training.
Original catalogue label: Beta. This label is retained for continuity and is not a current certification.
- 3.3
Model Asset Guard
Integrity and provenance checks for model-related files, manifests, and associated artifacts.
- Model integrity
- Provenance
Original project description and classification
Cryptographic verification of model weights, training data, and associated metadata to prevent tampering and ensure provenance.
Original catalogue label: Active. This label is retained for continuity and is not a current certification.
- 3.4
Post-Incident Proofs
Methods for deriving inspectable incident evidence from runtime records and recorded system behavior.
- Audit trails
- Forensics
Original project description and classification
An immutable, cryptographically sealed log of AI system behaviour, suitable for forensic analysis and compliance reporting.
Original catalogue label: Research. This label is retained for continuity and is not a current certification.
Test & data
- 4.1
PF Testbed
Test and evaluation environments for runtime mediation, observable decisions, and integration behavior.
- Testing
- Integration patterns
Original project description and classification
Testing framework for proof-carrying behaviour systems.
Original catalogue label: Active. This label is retained for continuity and is not a current certification.
- 4.2
Dataset Safety Specs
Formal descriptions of data lineage, admissibility conditions, and policy-related dataset properties.
- Data safety
- Compliance
Original project description and classification
Formal specifications for training-data safety and compliance.
Original catalogue label: Active. This label is retained for continuity and is not a current certification.
The organization directory includes additional repositories, maintenance work, and experimental projects. Individual repositories retain the authoritative implementation history.
Browse all repositories ↗