Security

Security

Mirrors SECURITY.md in the repo, plus the supply-chain verification story. If you remember one line: never report security problems in public issues, PRs, or discussions.

Reporting a vulnerability

Report privately through either channel:

1. GitHub Security Advisories (preferred) — this repo's Security → Report a vulnerability tab.
2. Email — security@algenta.ai.

Include where possible: impact description, the affected component (Python SDK or TypeScript SDK, and which module), reproduction steps or PoC, and the package version tested.

What to expect: acknowledgement within 3 business days, initial assessment and severity triage within 7, regular updates, coordinated disclosure with a published advisory once a fix is available, and credit (unless you prefer anonymity).

Supported versions

Channel	Supported
Latest release / main	✅
Older tagged releases	Best-effort — please upgrade to the latest

Scope

In scope — this repository's own code:

• Transport and auth handling in both SDKs (how credentials are sent, stored, or logged)
• The SDK's OAuth device-authorization client flow — polling, token storage, accidental token logging, local callback handling, validation, transport security
• Deserialization, injection, or other memory/logic-safety bugs in SDK code
• Insecure defaults in either package

Out of scope for this repository (redirect privately to security@algenta.ai):

• The private control-plane license-issuance service
• Algenta engine entitlement/license enforcement itself
• Any private activation or relay infrastructure

The trust boundary

The SDK is a plain HTTP client: it holds no license-signing keys, no entitlement-enforcement logic, and no secret shared with the engine. Every entitlement claim is independently verified and enforced by the closed engine. Forking the SDK, deleting every check in it, or replacing it with your own client does not change the execution capacity licensed to an Algenta engine — which is precisely why engine-side enforcement bugs belong to the engine's private channel, not this repo.

Supply-chain verification

Two independent signature stories cover what you install:

1. SDK releases (this repo). Every release is gated by an Ed25519-signed release-authorization record and ships Sigstore signature bundles plus SLSA provenance per asset. Full walkthrough and copy-paste verify commands: Versioning & Releases.

2. The native runtime wheels (proprietary data). algenta-runtime-native wheels carry a manifest.json with SHA-256 hashes of every artifact and a detached manifest.sig signature. The full Python client (pip install algenta) verifies the signature and every artifact hash against its embedded trust anchor before executing the runtime — fail-closed. The Mojo Quickstart example deliberately relies on the install path instead (pip/uv wheel-hash verification, pixi.lock version pinning) — treat it as a language example, not a hardened client.

Reporting cheatsheet

• Security vulnerability → private advisory or security@algenta.ai (never a public issue)
• Ordinary bug → GitHub Issues with the bug label
• Everything else → SUPPORT.md