Skip to content
Atlas

AI Disclosure

Last updated: 10 May 2026

Atlas is an AI-native operating system installed inside fund-backed portfolio companies. This page describes how AI is used in our marketing surface, how AI is used in customer engagements, and the operating discipline that governs both.

1. AI in the marketing surface

Some content on this Site — including portions of long-form positioning copy, parts of the privacy and terms boilerplate, and dynamic OG image generation — was drafted with the assistance of large-language-model platforms (Anthropic Claude, OpenAI GPT). All content is reviewed, revised, and approved by a human author before publication. The voice, the category-defining language, the pricing structure, and the legal posture are human-authored decisions that AI assists in expressing.

2. AI in Atlas customer engagements

When Perpetual Core engages a portfolio company under a signed Master Services Agreement, the install creates four operating layers inside the customer's environment: a substrate (data, connectors, memory), a department layer (Sales Ops, Finance Ops, Marketing/CS, optionally HR Ops or Industry-Specific Ops), an evaluation layer (the eval suite the CFO can audit), and a reporting layer.

Each department is built on large-language-model agents under the Claude Agent SDK and the OpenAI Agents SDK, wired through Anthropic's Model Context Protocol to the customer's connectors. The agents are not autonomous in the consumer sense. Every action that writes to a customer system is governed by an eval gate or a human-in-the-loop approval — Finance Ops in particular never auto-executes a write to the general ledger; every entry is human-approved during the install and for the foreseeable life of the engagement.

3. Customer data and foundation-model training

Customer data is not used to train any foundation model.

  • We do not send customer data to OpenAI or Anthropic for training. We use commercial inference APIs with the standard zero-data-retention posture available to enterprise customers.
  • We do not aggregate customer data across portfolio companies into a shared model. The cross-portfolio insight engine (used by Operating Partners to compare performance across the fund's portfolio) operates on sanitized, applicability-tagged metrics — not on raw customer data.
  • Customer data residency is per-portco isolated. Each engaged company has its own database tenant, its own connector secrets, and its own eval results. Cross-tenant queries are gated by Operating Partner authorization for that fund only.

4. Eval discipline

Every department installed by Atlas is held to evals authored during the install and run on a CI cadence afterward. Evals are versioned, replayable, and exportable. The CFO is given the full eval set before Finance Ops goes live; the eval results are part of the quarterly Operating Partner report. We treat evals as the auditable artifact of an AI-native operating system — the same way an external audit firm treats SOC 2 controls.

5. Human review and accountability

  • Every install has a named engagement lead at Perpetual Core
  • Every department write to a customer system is gated by an eval or by a human approval
  • The customer's Operating Partner or principal receives a weekly summary during the install and a quarterly review after go-live
  • Material changes to a department (new actions, new connectors, new eval thresholds) are reviewed with the customer before they ship

6. Disclosures during the buying process

If you submit an Atlas Discovery audit request through the Site, the audit itself is conducted with substantial AI assistance and reviewed by a human author before delivery. The Discovery deliverable identifies the AI-assisted sections explicitly. The same applies to install proposals and Statements of Work.

7. Updates

This disclosure is updated whenever the operating discipline changes materially. The "Last updated" date at the top of this page reflects the current version.

8. Contact

For questions about AI use, model providers, or customer-data handling:

Perpetual Core, Inc. Email: ai-disclosure@perpetualcore.com