plainpath
MethodDiscoveryBuildField notes
Get Started
MethodDiscoveryBuildField notes

Data Handling

Version 1.0 · Effective 8 September 2026

Sections

  1. 1Scope
  2. 2Summary of position
  3. 3What the audit collects
  4. 4Information that must not be submitted
  5. 5Where information is processed
  6. 6Access, encryption and controls
  7. 7Retention and deletion
  8. 8Service providers
  9. 9Discovery and the build
  10. 10Security incidents
  11. 11Your obligations
  12. 12Contact

1Scope

This document sets out what PlainPath collects during an engagement, where that information is processed, how it is protected, how long it is retained and what must not be submitted to us. It forms part of our Terms of Service.

It is written to be mapped against an internal AI standard or vendor assessment. Where a question is not answered here, we will answer it in writing on request.

2Summary of position

  • The audit is a structured conversation. It records descriptions of how work is performed. It does not collect customer records and it cannot accept file uploads.
  • The models used are Anthropic's Claude models, accessed through Anthropic's commercial API. Your data is not used to train those models.
  • Anthropic retains API inputs and outputs for up to 30 days for trust and safety review, after which they are deleted. They are not used for training.
  • PlainPath holds no SOC 2 certification. Our service providers hold SOC 2 Type 2 reports or equivalent attestations, described in section 8.
  • No data is transmitted to any other model provider.
  • Records are deleted on written request. Section 7 states the applicable timescales, including backups.

3What the audit collects

Compass, our audit agent, records the following:

  • The name, business email address and role of each participant.
  • The name of the engaging company.
  • The conversation as submitted by each participant.
  • A structured description of each process: its trigger, its steps, the role performing each step, the systems and data each step relies on, its frequency, its duration and the cost of failure.
  • The scores derived from that description, and the participant statements on which each score relies.
  • A summary of publicly available information about the company, compiled from a web search on the company name and its public website.

The audit does not collect the following:

  • File uploads. No upload facility exists and the audit endpoint accepts none.
  • System credentials. No login, key or token is requested at any point.
  • Customer, member or borrower records.
  • Access to any of your systems. The audit reads no system and connects to no system.

4Information that must not be submitted

The audit requires a description of work, not a sample of it. The following must not be entered:

Prohibited

  • Customer, member or borrower names, account numbers or balances
  • Social security numbers, taxpayer identification numbers or dates of birth
  • Loan files, financial statements or credit reports
  • Passwords, API keys or any other access credential
  • Any information your own policy classifies as restricted

A description such as "three figures are re-keyed from the client portal into a spreadsheet each month" is what the audit requires. The underlying documents are not.

Where restricted information is submitted in error, notify us and we will remove it from the record and from the next backup cycle.

5Where information is processed

Model processing. Conversation text is transmitted to Anthropic's commercial API to generate the next question, the process description, the scores and the written report. Anthropic is the only model provider that receives your data.

Under Anthropic's commercial terms, inputs and outputs submitted through the API are not used to train Anthropic's models. Reference: Anthropic Commercial Terms of Service and the Anthropic Privacy Center at privacy.anthropic.com.

Anthropic retains API inputs and outputs for up to 30 days and then deletes them. That period exists for trust and safety review and not for training. Zero Data Retention is not currently enabled on our Anthropic organization. Where an internal standard requires it, notify us during Discovery: it is an account-level arrangement with Anthropic and will be established before any build accesses your systems.

Application processing. The audit runs on software developed and operated by PlainPath, hosted on managed cloud infrastructure. Records are held in a single application database.

Backups. The database is backed up nightly to encrypted object storage. Seven daily, four weekly and three monthly copies are retained.

Correspondence. Report links and notifications are delivered by a transactional email provider. Replies are received through a managed email routing service.

6Access, encryption and controls

Authorized personnel. Administrative access is restricted to named PlainPath personnel, currently one individual. No shared accounts and no shared credentials are in use.

Authentication. Administrative access requires a hardware-backed passkey with biometric verification. Password authentication to the administrative interface is disabled in production. Automated access uses a single bearer credential held in the hosting environment and is present in neither the database nor the source code.

Session management. Each authenticated session is recorded with a seven-day expiry and a revocation flag. Revocation takes effect immediately and persists across restarts.

Encryption in transit. All requests are served over HTTPS with HSTS. A content security policy blocks inline event handlers, and framing by third-party sites is prohibited.

Encryption at rest. We rely on the volume-level encryption applied by our hosting provider. No additional application-layer encryption is applied to the database file. Backups are held in encrypted object storage.

Report access. Each report is served at a unique high-entropy URL. It is excluded from search indexing and cannot be enumerated. No password is required to open it. Where an internal standard requires authenticated access, we will apply it on request.

7Retention and deletion

Records are retained for the duration of the client or prospect relationship, as Discovery and any subsequent build rely on them. No automatic expiry is currently applied.

Deletion may be requested at any time in writing to [email protected]. On receipt:

  • The company record, all audit conversations, all process records and the report link are deleted from the live database within five business days. The report URL ceases to function.
  • Deleted records persist in backup copies already written until those copies expire, which takes up to three months on the schedule in section 5. Backups are not restored to recover deleted client data. We confirm in writing when the final copy has expired.
  • Correspondence already delivered remains in the recipient's mailbox and in ours.

8Service providers

We engage a small number of service providers to deliver the audit and subsequent engagements. Their functions are described in section 5: model processing, cloud hosting, encrypted backup storage, transactional email, public web search and payment processing.

Each provider is contractually bound to confidentiality and each holds a SOC 2 Type 2 report or an equivalent attestation. A current list identifying each provider, the data it processes and its attestations is available on request under the confidentiality terms of our engagement, or under a separate non-disclosure agreement where one is preferred.

PlainPath holds no SOC 2 certification and makes no representation to the contrary.

We do not sell data, we do not disclose it for advertising, and we do not use it to develop work for any other client.

9Discovery and the build

The audit is the least intrusive stage. Later stages require more, and each step is agreed in writing before it is taken.

Discovery. Sessions with your personnel, together with exports and samples you elect to provide. Where a sample is required to verify a figure, we request the smallest sufficient extract and ask that it be redacted where possible. Meeting recordings and transcripts are received only where you record them and provide them.

The build. Access to the systems the agent must reach, granted by you, scoped to the process, on accounts you control and may revoke. Read-only access is requested wherever read-only access is sufficient. Where a system provides no compliant means of access, we substitute a controlled interface and record the limitation in the build plan rather than circumventing your controls.

Model requirements. Where an internal policy requires a specific model provider, a specific processing region or a self-hosted model, notify us during Discovery. Such a requirement affects the build plan and the price and is accommodated as a matter of course.

10Security incidents

Where we become aware of unauthorized access to your data, we will notify you in writing within 72 hours, stating what is known, the action taken and our recommendation. Notification is not deferred pending a complete assessment.

11Your obligations

This document describes our obligations. It does not discharge yours under your own AI standard, your regulator's guidance, or your agreements with your funding bank. Where those requirements are more restrictive than this document, they govern, and we will operate within them. On receipt of the applicable standard we will state where we comply, where we do not, and what compliance would require.

12Contact

PlainPath

905 Ave K, Lubbock, TX 79401

[email protected]

See also: Terms of Service.

plainpath

The AI adoption practice of Primitive. We identify the work AI can run, build it, and train your team to own it.

start

The auditDiscoveryThe build

read

The methodField notesWho we are

talk

Contact[email protected]TermsData handling
© 2026 plainpath · a primitive company the A.G.E.N.T. Playbook was developed by dain studios and harvard data science review board members. read the article →