AI Disclosure

1. Why We Publish This

Several jurisdictions now require transparent disclosure when consumer services use generative AI (including California SB-942, the Colorado Artificial Intelligence Act, EU AI Act Art. 50, Utah AI Policy Act, and Texas TRAIGA). This page describes how Struxcor uses AI, what data is sent to AI providers, what the limitations are, and how to opt out.

2. AI Features in the Service

Struxcor uses large language models (LLMs) to power the following optional features. Each is invoked only at your explicit request (by clicking a button, sending a prompt, or uploading a document):

• Daily report narrative generation — drafts natural- language narratives from structured work, labor, equipment, and weather inputs.
• Voice-to-text transcription — converts field voice memos to text.
• Photo analysis — extracts description, condition, and tags from uploaded construction photos.
• Document OCR and extraction — recognizes text and structured fields from scanned submittals, RFIs, and pay applications.
• Compliance analysis — compares payroll records to prevailing wage determinations and flags potential discrepancies.
• Claims defense summaries — generates narratives and timelines from project records.
• Search and semantic retrieval — finds relevant documents and report entries using embeddings.

3. Model Providers and Data Handling

Our current AI provider is Anthropic, PBC, using the Claudefamily of models via the Anthropic API. Our relationship with Anthropic is covered by Anthropic's Commercial Terms of Service and Data Processing Addendum.

• What is sent to the AI provider: only the specific prompt and context needed to complete the request (e.g., the work items on a daily report, a single photo, a single document). We do not send your entire project database.
• No training on your data: Anthropic does not use Customer Data or prompts submitted through the API to train its foundation models.
• Retention at provider: Anthropic retains API inputs and outputs for up to 30 days for abuse-monitoring purposes, then deletes them. Flagged content may be retained longer where required by law.
• Location: inference runs on U.S.-based Anthropic infrastructure.
• Sub-processor status: Anthropic is listed as a sub-processor in our Privacy Policy and DPA.

We may add or switch AI providers in the future. We will update this page and notify account administrators at least 30 days in advance of a material change.

4. Limitations You Should Understand

• Hallucinations. LLMs can produce fluent, confident text that is factually wrong, including fabricated citations, rules, wage rates, calculations, dates, or events.
• Stale knowledge. Models have a training cutoff and may not reflect current regulations, prevailing-wage updates, or case law. Always verify against primary sources.
• Context limits. Long projects may exceed what the model can process in a single request; AI Outputs may summarize or omit information.
• Bias and errors in extraction. OCR, photo analysis, and transcription can misread handwriting, numbers, decimal points, and unit markings.
• Not a licensed professional. AI Outputs are not engineering opinions, legal advice, labor-law determinations, or certified inspection results. They do not satisfy licensure or certification requirements.
• Not a substitute for recordkeeping judgment. You remain the author and signatory of documents you submit; the AI is a drafting assistant.

5. Human Oversight Required

Before submitting AI-assisted content to a government agency, court, auditor, project owner, prime contractor, or other third party, a qualified individual (e.g., the Resident Engineer, Construction Manager, licensed engineer, or authorized representative) must:

• review the AI Output against the underlying records;
• verify numerical calculations (quantities, pay-item totals, wage math);
• confirm regulatory citations and wage determinations against primary sources (DIR, DOL, Caltrans);
• correct any errors; and
• sign off on the final submission.

The Service provides edit, diff, and audit tooling to support this review. You are the accountable party for anything submitted from your account.

6. Labeling of AI-Generated Content

Where feasible, we label AI-generated or AI-assisted content within the Service (for example, an "AI draft" badge on generated narratives). Labels are removed when a human author signs off. You are responsible for any downstream labeling required by law or contract when you distribute, file, or publish AI-assisted materials outside the Service.

7. Opt-Out and Controls

• Per-action: AI features are invoked only when you click a button or send a prompt. Ignoring AI buttons avoids all AI processing.
• Organization-wide: administrators can disable AI features for their entire organization in Settings → Organization → AI Features.
• Per-project: AI can be disabled on sensitive projects via the project settings page.
• Request-level delete: AI prompts and outputs are associated with your account and can be deleted via the Service or by contacting privacy@struxcor.com.

8. Automated Decision-Making

We do not use AI to make automated decisions that produce legal or similarly significant effects on individuals (for example, employment, credit, housing, insurance, or access to public services). AI features are drafting and analysis tools used under human control.

9. Intellectual Property in AI Outputs

• As between you and us, you own the AI Outputs generated from your prompts and Customer Data.
• AI Outputs may be similar or identical to outputs generated for other users from similar prompts. We cannot guarantee exclusivity.
• Copyright protection for purely AI-generated works is unsettled in many jurisdictions (e.g., the U.S. Copyright Office generally requires human authorship). Substantive human contribution is advisable if copyright is important to your use case.

10. Security of AI Data Flows

• All traffic between your browser, our servers, and Anthropic is protected with TLS 1.2+.
• Prompts are authenticated against your account; API keys to Anthropic are held server-side and never exposed to clients.
• We log prompt metadata (user, timestamp, token counts, model) but redact known sensitive-field patterns from long-term analytics.

11. Children

AI features are not designed for or directed to individuals under 18 and must not be used to process personal data about minors without a lawful basis.

12. Changes

We will update this Disclosure when models, providers, retention terms, or available controls change. The "Last updated" date and version will change accordingly.

13. Contact