AI Disclosure

Last updated: 7/16/2026

MoldMind is an AI-assisted inspection report platform. This disclosure explains how artificial intelligence is used in the Service, which third-party AI providers we rely on, what the AI can and cannot reliably do, and how the inspector remains the decision-maker for every Report.

1. The AI-Assisted, Not AI-Generated, Principle

Every Report produced through the Service is reviewed and explicitly approved by the subscribing Inspector before it is finalized or delivered. The Service does not auto-send Reports, does not auto-approve drafts, and does not produce binding professional conclusions on its own. AI is a drafting and structuring aid; the Inspector is the professional of record and the decision-maker for every finding, every recommendation, and every signed Report.

Anywhere AI-generated narrative or AI-extracted data appears in the Service or in a Report, the user is the professional accountable for verifying accuracy, completeness, and fitness for the inspection context. Notwithstanding any provision regarding ownership or use of Output in the Terms of Service, you remain solely responsible for the content and use of any finalized Report.

2. What the AI Does in MoldMind

The Service uses AI for the following well-defined tasks:

• Report narrative drafting. AI generates the prose body of inspection Reports — Assessment Report, Remediation Protocol, and Client Summary Letter — from the structured field data you upload and the standards encoded in our prompts (IICRC S520, IICRC S500, AIHA, ACGIH, ASHRAE 160, ASTM D7391, and similar standards where applicable). The Inspector reviews and edits the draft before approval.
• Photo classification and grouping. AI groups uploaded photos by room, location, and finding type, and can read visible meter displays and similar visual indicators. The Inspector confirms the classifications before they are written into the structured Report fields.
• Voice memo transcription and reformatting. AI transcribes voice memos via a third-party speech-to-text vendor and then reformats the transcript into professional inspection language. The Inspector reviews the transcript and the reformatted output.
• Document parsing. AI extracts structured data from photos of handwritten chain-of-custody forms, lab PDFs, and similar documents using vision-capable models. The Inspector verifies extracted fields against the original document before approval.
• Correction-pair learning. When the Inspector edits AI output, the edit pair (AI draft → Inspector correction) is captured and used to bias future drafts toward the Inspector's preferred style, terminology, and findings. This learning is per-Inspector and per-organization; we may also use anonymized correction pairs in aggregate to improve the Service generally, consistent with the Privacy Policy and the Aggregated and De-identified Data section of the Terms.

The Service does not use AI to: send communications to third parties on the Inspector's behalf without approval; finalize Reports without explicit approval; make consequential decisions (such as whether to remediate a property or condemn a building); or make decisions about access to the Service.

3. AI Vendors

We may add, change, or remove AI vendors as the Service evolves. The current list of all sub-processors (including AI vendors) is published at /subprocessors and changes follow the material-change notification process described in the Terms and Privacy Policy.

4. Training-Data Posture

We do not send Customer Content (in identifiable form) to any third party to train that third party's foundation AI models. Customer Content is sent to our AI vendors to generate Output for the Customer, under vendor agreements that prohibit training on it (see "Vendor data retention" below).

We do, however, use Customer Content for the following service-improvement purposes consistent with the Privacy Policy:

• Per-Inspector and per-organization correction-pair learning. When an Inspector edits AI output, the (AI draft → Inspector correction) pair is captured and used to bias future drafts toward the Inspector's preferred style, terminology, and findings. This learning is scoped to the relevant Inspector and organization and is not used to train third-party foundation models.
• Aggregated, de-identified data. We create aggregated and de-identified data derived from your use of the Service. As described in the Privacy Policy and in Section 8 of the Terms of Service, once data is de-identified it is no longer Customer Content or personal data, is owned by MoldMind, is retained perpetually, and may be used for any lawful purpose — including developing, training, and improving models and artificial-intelligence systems (whether or not related to the Service) and creating, licensing, or selling standalone products. The no-training commitments above and our vendor agreements apply to identifiable Customer Content, not to de-identified data.

Vendor data retention. Our AI vendors have their own policies on whether they retain or use customer-submitted content for vendor model improvement. Our current vendor posture is:

• Deepgram. We send mip_opt_out=true on every transcription request, instructing Deepgram not to retain audio or transcripts for vendor model improvement.
• Anthropic. Our AI vendor agreements with Anthropic generally prohibit training of Anthropic's models on Customer Content sent through the API.
• OpenAI. We use OpenAI only to generate text embeddings (vector representations used for similarity search), accessed through the Vercel AI Gateway. OpenAI's API platform terms provide that content submitted through the API is not used to train OpenAI's models by default, and we do not enable any such training. Only short text snippets derived from correction pairs and support questions are embedded; we do not send photos or audio to OpenAI.
• Vercel (hosting and AI gateway). Vercel's commercial terms govern the inference traffic it proxies; we may, at our discretion, enable or disable zero-data-retention configurations subject to vendor plan tiers and our operational posture, and the current configuration may change over time.

We monitor our vendor agreements and may change these vendor configurations as our vendor relationships, plan tiers, and operational requirements evolve. We will update this Disclosure if vendor practices materially change.

5. Limitations of AI Output

AI Output is probabilistic. You acknowledge and agree, and must communicate to anyone relying on a Report, that:

• AI may produce inaccurate, incomplete, or fabricated content. This includes invented citations, misclassified photos, missed findings, mistranscribed voice memos, and incorrect extractions from lab or chain-of-custody documents.
• AI may apply standards incorrectly. The Service uses prompts to bias outputs toward specific industry standards, but the AI may misapply, conflate, or partially apply those standards. The Inspector is responsible for verifying that any standards citation in a Report is accurate and applicable to the inspection context.
• AI output is not a substitute for professional judgment. The Service is not a substitute for an Inspector's training, certification, experience, on-site observation, or independent verification of laboratory results.
• AI output is not legal, medical, environmental, or other professional advice from MoldMind. Whether an Inspector's use of the AI Output is consistent with applicable professional, regulatory, or licensing requirements is solely the Inspector's responsibility.

You agree not to rely on AI Output without the independent professional review required by the Service's design. See also the Warranty Disclaimer and Limitation of Liability in the Terms of Service.

6. Notice to End Recipients of Reports

If you deliver a Report generated by the Service to a third party (a client, homeowner, contractor, insurer, attorney, or other recipient), you should communicate the AI-assisted nature of the Report to that recipient where relevant. The AI-assisted nature of the Service is disclosed through this AI Disclosure and through the acceptance flow you complete when you use the Service. Each finalized Report carries a signed Inspector Certification in which the subscribing Inspector certifies that the Report reflects the Inspector's professional opinion, establishing the Report as the Inspector's professional work product. The Report PDF does not itself print an "AI-assisted" label, so where disclosure of AI assistance to a recipient is relevant, you are responsible for communicating it.

7. Aggregation for Improvement

We may aggregate de-identified inspection data across our customer base to improve the Service and produce aggregated insights (such as industry reports, regional trends, and benchmarks). Aggregation is performed in a manner intended to prevent re-identification of individual customers, Inspectors, or properties. We do not commit to deliver any specific aggregation product, and any aggregation we perform may change over time. See the Privacy Policy and Terms of Service for additional detail.

8. Regulatory Disclosures

8.1 Federal (United States)

There is no comprehensive federal AI regulation in the United States at this time. To the extent any federal AI rule, executive order, or sector-specific regulation applies to the Service, we will update this disclosure as the regulatory landscape evolves.

8.2 California

• AB 2013 (training-data disclosure). MoldMind does not itself develop generative AI foundation models. We use third-party foundation models (Anthropic Claude, Deepgram speech-to-text, and OpenAI text-embedding models). The model developers publish their own training-data disclosures. We do not train models on Customer Content (see Section 4).
• SB 1001 (bot disclosure). The AI-assisted nature of the Service is disclosed to Inspectors through this AI Disclosure and the acceptance flow, and each finalized Report carries a signed Inspector Certification establishing it as the Inspector's professional work product. Any customer-facing chat-style interaction provided by the Service that involves AI is identified as AI-assisted.
• California Generative AI Accountability Act (SB 896) and related developments. We monitor California AI legislation and will update this disclosure as additional obligations come into effect.

8.3 Colorado

The Colorado AI Act framework (originally SB 24-205, currently being reshaped under SB 189 and subsequent legislative action) regulates "high-risk" AI decision systems that make or substantially contribute to "consequential decisions" affecting consumers. The Service produces an Inspector-assistive Output that is reviewed and approved by a human professional before any decision is communicated; the human Inspector is the decision-maker. We do not consider the Service to be a "high-risk AI system" or to produce "consequential decisions" within the current scope of the Colorado AI Act. MoldMind monitors Colorado AI legislation and will update this disclosure as the regulatory framework stabilizes. Colorado customers requiring additional disclosures should contact legal@moldmindai.com.

8.4 Connecticut

Connecticut Public Act 25-44, effective July 1, 2026, addresses AI-driven consequential-decision systems among other matters. As described above, the Service is human-in-the-loop and does not produce consequential decisions through fully automated means. We will update this disclosure if Connecticut's regulatory framework imposes additional obligations applicable to the Service.

8.5 European Union — EU AI Act

The EU AI Act applies to providers and deployers of AI systems placed on the market or put into service in the European Union. MoldMind is US-only at launch. If we expand to offer the Service in the EU, this disclosure will be updated to align with the AI Act's risk-tier classification, transparency, and documentation requirements, and we will provide the disclosures required of "providers" or "deployers" as applicable to our role.

8.6 Other jurisdictions

We monitor AI regulation in the United Kingdom, Canada, Australia, Brazil, and other jurisdictions, and will update this disclosure as we expand into those jurisdictions or as obligations otherwise become applicable.

9. Security and Prompt-Injection Protections

The Service applies a layered defense against prompt-injection and AI-manipulation attempts, including:

• Input sanitization that wraps user-controllable text in delimited tags before it is sent to the model;
• Instruction-hierarchy guards in our system prompts that distinguish trusted system instructions from untrusted user content;
• Output validation that asserts a minimum content structure before a draft is rendered;
• Per-call cost caps that prevent runaway AI usage; and
• Logging and rate limits that surface unusual patterns for security review.

These protections are part of our acceptable-use enforcement and do not change the Inspector's responsibility to review every Output before approval.

10. Changes to This Disclosure

We may update this AI Disclosure from time to time. For minor or clarifying changes, we will post the revised version with a new "Last updated" date. For material changes, we will provide notice consistent with the Terms of Service (Section 20) and the Privacy Policy (Section 18).

11. Contact

Questions about AI use in MoldMind:

MoldMind LLC 2389 Main Street, Suite 100 Glastonbury, CT 06033 United States

ai-disclosure@moldmindai.com