AI & Machine Learning Disclosure

1. Scope

This disclosure explains how TheAccessible.org uses artificial-intelligence and machine-learning systems ("AI") when operating the Service. It supplements our Privacy Policy and should be read together with it.

"AI" here means any model used to generate, analyze, classify, or transform content — including large language models, vision models, and speech models — regardless of whether the model is ours or a third party's.

2. Where we use AI

We use AI to assist with:

• Document analysis — identifying structure, reading order, headings, tables, lists, alt-text candidates, and tagging needs in submitted PDFs.
• Remediation suggestions — proposing alt text, table summaries, and accessible markup that a human author or reviewer then accepts, edits, or rejects.
• Content generation within authoring tools — drafting outreach emails, summaries, and similar text when a user explicitly asks for it.
• Classification and bulk processing — triaging low-risk tasks such as routing and prioritization.

We do not use AI to make hiring, credit, legal, or similar decisions about users.

3. Models and providers

Provider and model selection may change as better options become available. The current list is kept in sync with our Subprocessors list.

4. What data is sent to AI providers

When an AI-assisted feature runs, we send the provider only the data needed to complete that task:

• Document-analysis tasks — the page or region being analyzed, plus minimal context (e.g. surrounding pages needed to establish reading order).
• Remediation suggestions — the element under review and a short context window.
• Authoring assistance — the prompt you supply and any content you explicitly attach.

We do not send unrelated customer content, account credentials, payment data, or data from other customers' accounts.

5. Training — we do not allow it

We do not use Customer Content to train AI models, and our agreements with Anthropic and Google prohibit them from training their models on content submitted through their enterprise APIs. This commitment is enforced contractually, not only by configuration flags, and it applies whether or not you use the Service's AI-assisted features.

6. Provider retention

• Anthropic — API inputs and outputs are retained by Anthropic only as needed to provide the service, detect abuse, and meet legal obligations, and are not used for training. See Anthropic's current commercial terms for specifics.
• Google (Gemini) — under the Gemini API enterprise terms, prompts and responses are not logged for training and are retained only as needed to operate and secure the service.

We do not retain a copy of a prompt at the AI provider longer than needed for the Service to return a response and for minimal troubleshooting. For our own retention of inputs and outputs, see the Data Retention Policy.

7. Human review and automated decision-making

AI-generated output is a suggestion — it is surfaced to a human (you, your reviewer, or our staff on support tasks) who decides whether to accept, edit, or reject it. We do not make solely-automated decisions that produce legal or similarly significant effects about users.

If we ever introduce a feature that produces such decisions automatically, this disclosure will be updated and affected users will be notified.

8. Accuracy and limitations

AI systems can be confidently wrong. We test AI-assisted features for quality and safety, but we do not guarantee that output will be accurate, complete, or free of bias in every case. Remediated output should be reviewed by a person familiar with the document's context and with WCAG requirements before being relied upon for compliance.

The Service is a tool to assist accessibility work; it does not certify compliance with any particular standard.

9. Data residency

AI processing occurs in the regions listed in the table in section 3 (currently the United States). Where personal data is transferred internationally, we rely on the safeguards described in the Privacy Policy, section 7.

10. Your choices

You can control AI use as follows:

• Disable AI services — toggle Disable AI services in Settings → Privacy & AI (stored as profiles.disable_ai_services). When enabled, AI-dependent endpoints (document conversion, HTML/DOCX/PPTX remediation, form conversion, audio transcription, exemption review, org-chart vision) refuse the request with AI_DISABLED_BY_USER before any provider call. No content from your account is sent to Anthropic, Google, or OpenAI under any flow while this setting is on. Changes to this setting are recorded in your account audit log.
• Use API-only mode — customers using our API can omit the AI-assisted endpoints and call only deterministic conversion routes.
• Bring-your-own-key (BYOK) — Some service tiers allow the customer to use their own keys for AI services. Contact sales@theaccessible.org for more information.

You can exercise your data-subject rights (including deletion of prompts we have logged) via the DSAR Procedure.

11. Changes to this disclosure

We will update this disclosure when we add or change AI providers, expand AI-assisted features, or change how we handle data sent to providers. For material changes we give at least 30 days' advance notice in line with our Privacy Policy. Prior versions remain available from the version history link below.

12. Contact

Questions about AI use or this disclosure: privacy@theaccessible.org.