Skip to content
@invariant-systems-ai

Invariant Systems

Verifiable computing infrastructure for the AI era. Makers of AIIR — AI Integrity Receipts.

Invariant Systems

Verifiable computing: receipts, proofs, and replayable evidence.


The Problem

Modern systems make claims constantly: this commit declared AI involvement, this model produced that output, this action earned that payout. Almost none of those claims come with evidence a third party can check. The most concrete case today: AI tools write an increasing share of production code, but git history can't answer which changes were AI-generated. Trailers are inconsistent, easy to strip, and not machine-verifiable.

What We Build

AIIR (AI Integrity Receipts) is our first shipped product. It generates deterministic, content-addressed receipts for commits with declared AI involvement and verifies them anywhere: locally, in CI, or offline, without trusting a central service.

Public research extends the same receipt model further: inference receipts and receipted actions (attestable AI), plus evidence-first capsules in physics and mathematics, published Zenodo-first with open DOIs and explicit claim boundaries. See the Research page for the linked Zenodo records and reproducibility capsules.

AIIR | AI Integrity Receipts

PyPI pip install aiir
GitHub Action invariant-systems-ai/aiir@v1
GitLab CI One include: line: push and MR receipts; GitLab Technology Partner
VS Code Extension: local-first commit receipts, inline verification, receipt explorer
AI Assistants via MCP Works with Claude, Copilot, Cursor, Continue, Cline, Windsurf
License Apache 2.0; zero runtime dependencies (CLI core); Python 3.9+

Security posture: 2,499 collected tests, 100% coverage, and a public threat model with 150+ documented security controls, plus ClusterFuzzLite fuzzing, mutation testing, and conformance vectors. JSON and deterministic CBOR receipt formats. Optional Sigstore signing and PEP 740 attestations.

How it fits: AIIR fills the authorship-provenance gap before build-level tools (SLSA, in-toto, SCITT) kick in. See ecosystem positioning and the public Research page for the current public picture.

Detection is heuristic: AIIR records what is declared: Co-authored-by trailers, bot authors, and AI-tool markers. Agent-mode sessions such as Copilot Chat, Claude Code, and Cursor Agent do not add these markers today, so those commits can still land in the human bucket unless the tool or user declares them explicitly. See detection scope for current public limits.

Company

Invariant Systems, Inc. is a Delaware C-Corp founded in 2025.

Website | Docs | Research | Browser Verifier | Contact: noah@invariantsystems.io


© 2025-2026 Invariant Systems, Inc.

Popular repositories Loading

  1. .github .github Public

    Organization profile

  2. research-evidence-log research-evidence-log Public

    Public research evidence log for evaluation plans, evidence packages, and paper-facing capsule mirrors.

    Python

  3. mub6-wall-atlas mub6-wall-atlas Public

    An exact, machine-checkable defect atlas for the MUB(6) feasibility wall: max f = (88 + 3 sqrt6)/100 for the canonical product triple. Reproduce, audit, formalize, or break scoped bricks.

    Python

Repositories

Showing 3 of 3 repositories

People

This organization has no public members. You must be a member to see who’s a part of this organization.

Top languages

Loading…

Most used topics

Loading…