AI, transparency & the law
Last reviewed: July 2026
AI is now regulated by a growing patchwork of federal and California law. This page sets out, plainly, where ADI-AI and Scribe50 stand against every AI-related rule we are aware of — the ones that apply to us, the many that do not, and why. We would rather show you the whole map than a marketing badge.
The short version: nearly all of these laws attach only to very large AI operators — systems with more than a million users, platforms with millions more, or frontier developers with hundreds of millions in revenue. Scribe50 sits far below those thresholds, so their obligations do not legally bind it. We built it to align with their intent anyway. Alignment with the intent of a law is not the same as being certified against it, and we do not claim any certification we have not earned.
Why our design answers most of these laws before they ask
Most AI regulation exists to solve problems that come from AI running on someone else's servers at massive scale: opaque cloud models, harvested data, undisclosed synthetic content, automated decisions made about people. Scribe50 is built the opposite way, and that architecture is what keeps it clear of — and aligned with — these rules:
- It runs entirely on your device. Transcription happens locally; there is no cloud AI service behind it and no server-side copy of your audio or transcript.
- It sends no telemetry and needs no account. Scribe50 FREE has no licensing component at all — no key, no sign-in, no validation call.
- It uses a pre-trained speech model as-is. Scribe50 bundles an established open speech-to-text model rather than training one, which is central to the one California rule below that has no size threshold.
- It labels its AI output and lets you report it. Transcripts carry a clear "generated by AI, may contain errors" notice, and a Report an Issue control lets you flag inappropriate or incorrect AI output directly to us.
United States — federal
There is no comprehensive federal AI statute in force as of July 2026. The federal items that touch AI are voluntary frameworks, agency-facing executive actions, or general laws that reach every business. Only one — the FTC Act's ban on deceptive claims — creates a live obligation for a company our size, and it is simply a duty to tell the truth.
| Law / framework | What it governs | Where Scribe50 stands |
|---|---|---|
| NIST AI Risk Management FrameworkAI RMF 1.0 + Generative AI Profile (voluntary) | A voluntary structure for identifying and managing AI risk across its lifecycle. Not law. | Voluntary — we align We use its principles as a design guide. There is nothing to "comply" with. |
| FTC Act § 515 U.S.C. § 45 — deceptive practices | Prohibits false or unsubstantiated claims to consumers, including overstated AI accuracy, privacy, or compliance claims. | Applies — as honesty It binds us the way it binds everyone: we only make claims we can stand behind. It is why this page is worded the way it is. |
| TAKE IT DOWN ActPub. L. 119-12 (2025) | Criminalizes non-consensual intimate imagery and requires user-content platforms to run a fast notice-and-takedown process. | Not applicable Scribe50 hosts and publishes no user content and is not a covered platform. |
| Executive actions on AIEO 14179 (2025), EO 14365 (2025) | Direct federal agencies on national AI policy and on the federal government's own use and procurement of AI. | Not applicable These direct government agencies, not independent software developers. They impose no obligation on Scribe50. |
| Section 508 & COPPA29 U.S.C. § 794d; 15 U.S.C. § 6501 | Federal-procurement accessibility, and online collection of data from children under 13, respectively. | Not applicable We sell to no federal agency and collect no data from children (or, in FREE, any data at all). We treat accessibility as good practice regardless. |
California
California has the most developed AI-transparency law in the country. Every size-gated California statute sets its bar far above where Scribe50 operates, so those obligations do not bind it. We list them in full anyway, and we single out the one rule that carries no size threshold.
| Law | What it governs & who it covers | Where Scribe50 stands |
|---|---|---|
| California AI Transparency ActSB 942 (2024), amended by AB 853 (2025) | Requires generative-AI providers to offer a free AI-detection tool and embed provenance/watermarks. Covers providers with over 1,000,000 monthly users. | Below threshold Scribe50 is far under a million monthly users, so its obligations do not apply. Our AI-output labelling reflects the same goal. |
| Large-platform & capture-device dutiesAB 853 (2025) | Provenance-detection duties for large online platforms (2,000,000+ users) and, later, for hardware capture-device manufacturers. | Below / not applicable We are not a multi-million-user content platform, nor a device manufacturer. |
| Generative AI: Training-Data TransparencyAB 2013 (2024) — in force Jan 2026 | Requires a developer of a generative-AI system to publish a summary of the data used to train it. Has no revenue or user-count threshold. | The one we watch Scribe50 bundles a pre-trained speech model as-is, so the training-data disclosure duty rests with that model's original developer, not us. See the note below. |
| Transparency in Frontier AI ActSB 53 (2025) | Safety-framework, incident-reporting, and whistleblower duties for frontier-model developers. Covers models trained above 1026 FLOPs; heaviest duties above $500M revenue. | Below threshold Scribe50 trains no frontier model and is orders of magnitude below both the compute and the revenue bar. |
| Bot Disclosure ActSB 1001 (2018) | Requires disclosing that a "bot" is not human when it messages people to drive a sale or influence a vote. | Not applicable Scribe50 is a transcription tool, not a bot that converses with people to sell or persuade. |
| Companion Chatbot safetySB 243 (2025) | Safety, disclosure, and minor-protection duties for AI "companion chatbots." | Not applicable The law expressly excludes productivity tools. Scribe50 has no conversational companion features. |
| CCPA / CPRA & automated-decision rulesincl. AB 1008 (2024); CPPA ADMT regs | Privacy and automated-decision-making duties for "businesses" over $25M revenue or handling 100,000+ people's data, making significant decisions about people. | Below threshold We are under the business thresholds, hold essentially none of your data, and make no automated decisions about anyone. |
| Generative AI Accountability ActSB 896 (2024) | Directs California state agencies to assess and disclose their own use of generative AI. | Not applicable It governs the state government's use of AI, not private software. |
The one rule without a size threshold: AB 2013
Almost every law above lets us off by size. California's AB 2013 does not — it asks any developer of a generative-AI system to publish a summary of its training data, no matter how small the company.
The word that matters is "developer." AB 2013 places the duty on whoever trains or substantially modifies the AI model. Scribe50 bundles an established, pre-trained open speech-to-text model and runs it as-is; we do not train it or alter its weights. On that basis the training-data-summary duty sits with the model's original authors, not with us. If that ever changes — if we fine-tune or substantially modify a model ourselves — we will publish a training-data summary consistent with AB 2013 and say so here. We would rather tell you our reasoning than assert a conclusion.
What we do not claim
We are not certified against any of these laws, and we do not say we are.
Being below a law's coverage threshold means its obligations do not apply to us — it does not mean a regulator has reviewed and blessed us, and we will not imply otherwise. Where we say we "align" with a rule's intent, we mean our design choices — local processing, no telemetry, labelled AI output, an easy way to report problems — serve the same purpose the rule was written to serve. That is a design decision we can show you, not a compliance certificate we are borrowing. This page is our current, good-faith reading of how these laws apply to a product like ours; it is not legal advice, and the law in this area is changing quickly.
Related
Security — how Scribe50 is built and secured. Privacy Statement — what personal information we handle and why. Terms of Service — licensing, billing, and the limits of what the Software guarantees.
Questions about anything on this page, or want to report an issue with AI-generated output? Write to us at support@adi-ai.ai.