Dictation Privacy Hub: The Complete Guide to Protecting Your Voice Data

TL;DR: Your voice is biometric data — a permanent identifier as unique as your fingerprint that cannot be changed after a breach. Cloud dictation apps send this data to remote servers where it can be stored, shared, breached, or used for AI training. On-device dictation keeps all audio on your Mac, eliminating server-side exposure. This hub organizes our complete library of dictation privacy guides to help you protect your voice data.

In 2025, Google agreed to a $1.375 billion settlement with Texas for unlawfully collecting biometric data including voiceprints. Apple paid $95 million to settle a Siri recording lawsuit. Amazon eliminated the option to store Echo recordings locally. The message is clear: voice data is a high-value target, and the companies you trust with your voice do not always protect it.

Whether you are a healthcare professional bound by HIPAA, a lawyer protecting attorney-client privilege, or simply someone who values privacy, understanding how dictation tools handle your voice is essential. Explore the guides below to find exactly what you need.

Disclosure: Voibe is our product. We compare fairly and acknowledge competitor strengths throughout our guides.

Voice data privacy has reached an inflection point. Three trends are reshaping how dictation tools handle your audio:

Regulatory enforcement is accelerating. Over 107 BIPA class-action lawsuits were filed in Illinois in 2025 alone, targeting companies that collected voiceprints without consent. The Clearview AI settlement reached $51.75 million. GDPR classifies voice recordings as special-category biometric data requiring explicit consent. HIPAA violations involving voice data carry fines up to $2.07 million per violation category per year. On-device processing sidesteps all of this regulatory complexity by ensuring no voice data is collected in the first place.

Big tech is collecting more, not less. Amazon eliminated its local-only voice processing option in March 2025, requiring all Echo recordings to travel to the cloud. A University of Washington study found Alexa data is shared with up to 41 advertising partners. The FTC fined Amazon $25 million for keeping children's voice recordings indefinitely after parents requested deletion. Wispr Flow, positioned as a productivity tool, faced a viral privacy backlash when users discovered it captures screenshots of the active window every few seconds and sends them to external servers (OpenAI and Meta) for context awareness — with no offline alternative. The company reportedly banned the user who first raised these concerns publicly, and only updated its policies after significant public pressure.

"On-device" is not always private. Superwhisper processes speech locally but saves audio recordings by default — users have repeatedly requested the ability to disable this on the public feedback board, with no resolution. API keys are stored in plaintext JSON on disk. A persistent microphone indicator stays on between dictations. These behaviors create privacy risk even when the core transcription is on-device. The architectural difference between "runs locally" and "keeps all data under your control" matters.

On-device AI has closed the accuracy gap. OpenAI's Whisper large-v3 achieves a 2.7% word error rate on clean English audio — competitive with cloud services. Apple Silicon's Neural Engine enables real-time local inference. Tools like Voibe now deliver cloud-quality accuracy with zero data leaving your device.

These trends mean the choice between cloud and on-device dictation is no longer a trade-off between accuracy and privacy — it is purely a privacy decision. And within on-device tools, architecture and data-handling defaults matter as much as where transcription happens.

Each guide below covers a specific aspect of dictation privacy in depth. Start with whichever topic is most relevant to your situation.

HIPAA Dictation: Requirements, Tools, and Compliance Guide

Healthcare professionals who dictate patient notes handle Protected Health Information (PHI). Voiceprints are explicitly listed as HIPAA identifier #16, meaning dictation audio is inherently PHI. This guide covers the five HIPAA requirements for dictation software, compares tool compliance (Dragon Medical One at $79-99/mo vs. Voibe at $149 lifetime vs. Superwhisper at $249.99 lifetime), penalty structures up to $2.07M per category, and implementation checklists.

Read this if: You work in healthcare, handle patient data, or need to understand HIPAA dictation compliance.

See also: Dragon Medical Alternatives for Mac — a comparison of 7 Dragon Medical One alternatives including on-device options that keep patient audio off cloud servers.

Voice Data Privacy: How Dictation Apps Collect, Store, and Use Your Audio

Cloud dictation apps collect five categories of data from your voice: raw audio, transcripts, biometric voiceprints, metadata, and background audio. This guide explains exactly what each dictation service collects, how data is shared with third parties (Alexa shares with up to 41 ad partners), the regulatory frameworks that protect you (GDPR, BIPA, CCPA), and how to minimize exposure.

Read this if: You want to understand what happens to your voice data after you speak into a dictation app.

Cloud vs. Local Dictation: Privacy, Speed, and Accuracy Compared

The fundamental choice in dictation privacy is where your audio gets processed — on remote servers or on your device. This guide provides a technical comparison across privacy, latency (100-500ms cloud overhead vs. near-zero local), accuracy (Whisper large-v3 at 2.7% WER matches cloud services), and cost (Voibe lifetime at $149 vs. Otter Pro 3-year at $611.64, vs. Superwhisper lifetime at $249.99).

Read this if: You want a data-driven comparison to decide between cloud and on-device dictation.

How Whisper Works: OpenAI's Speech Model Explained for Mac Users

Whisper is the open-source AI model that enables private on-device dictation. Trained on 1 million+ hours of audio, it runs locally on Apple Silicon's Neural Engine with Core ML delivering 3x faster inference than CPU-only. This guide explains the encoder-decoder architecture, model sizes from tiny (39M params) to large (1.55B), and why Apple Silicon makes real-time local speech recognition possible.

Read this if: You want to understand the technology behind on-device dictation and how Apple Silicon enables it.

Apple Dictation Privacy: What Data Apple Collects and How to Stop It

Apple Dictation is free and mostly on-device on Apple Silicon, but has privacy caveats. The "Improve Siri & Dictation" setting sends audio samples to Apple. Apple paid $95M in January 2025 to settle a Siri recording lawsuit. This guide covers exactly what Apple collects, step-by-step instructions to disable data sharing, HIPAA limitations, and how Apple Dictation compares to fully on-device alternatives.

Read this if: You use Apple's built-in dictation and want to maximize its privacy settings.

Companion piece: Apple Dictation Pricing Breakdown — the dollar-cost analysis on what "free" actually costs in time, accuracy losses, and HIPAA exposure (Apple does not sign BAAs, which makes Apple Dictation a regulatory blocker for any PHI workflow).

Offline Dictation Privacy on Mac: How On-Device Speech to Text Keeps Your Data Safe

Our comprehensive deep-dive into the cloud dictation data pipeline, the specific risks at each stage (transmission, server processing, retention, training use), which Mac professionals face the highest risk, and a detailed privacy comparison of every major Mac dictation tool. Includes a verification checklist and decision framework.

Read this if: You want the most thorough analysis of Mac dictation privacy with tool-by-tool comparisons.

Typeless Privacy Issues: What Researchers Found and Why Cloud Dictation Is Risky

A real-world case study of the gap between "privacy-first" marketing and cloud-based architecture. Typeless markets "on-device history" and "zero data retention," but its own privacy policy confirms audio is processed on cloud servers, and a November 2025 reverse-engineering analysis reported routing to AWS us-east-2 alongside URL capture, window-title collection via the accessibility API, and broad permission requests. Introduces the 8-point Dictation Privacy Audit framework you can apply to any dictation app before granting microphone access.

Read this if: You want to see what happens when cloud dictation marketing does not match architecture — or you need a framework to evaluate any dictation app's privacy claims.

Is Wispr Flow Safe? Privacy, Delve Audit Scandal & Verdict (2026)

A current-state safety investigation of Wispr Flow. Walks through Wispr's actual cloud architecture (audio processed by Baseten, text by OpenAI/Anthropic/Cerebras, storage in AWS us-east-1), the Privacy Mode mechanics (off by default for non-HIPAA users; locks irreversibly when a BAA is signed), and the March 2026 Delve compliance scandal — Wispr Flow's prior compliance vendor was named in a credible fake-audit investigation that 99.8% of 494 SOC 2 reports shared identical boilerplate text. Wispr Flow has remediated transparently with A-LIGN as the new auditor, Drata as the new compliance platform, and SafeBase for the trust center. Includes a five-question Wispr Flow Safety Decision Tree.

Read this if: You currently use or are evaluating Wispr Flow, especially for sensitive or regulated work — or you want to understand how the Delve compliance scandal changes the trust calculation for cloud SaaS dictation.

Is Superwhisper Safe? Privacy Modes, Local Recordings & Verdict (2026)

A current-state safety investigation of Superwhisper. Walks through the architectural split between on-device modes (Tiny / Base / Small / Standard Whisper / Parakeet — genuinely local) and cloud modes (Ultra transcription + Super Mode LLM post-processing — proxied through Superwhisper to OpenAI / Anthropic / Google / Groq / Meta / Mistral / Grok). Documents three structural caveats: local audio recordings are ON by default (23 votes on the public feedback board to make it opt-in), API keys for cloud-mode providers are stored in plaintext JSON on disk (15+ votes), and the privacy policy was last updated June 19, 2024 and does not separately describe cloud-mode handling. Includes a five-question Superwhisper Safety Decision Tree and a Superwhisper Safety Audit checklist.

Read this if: You currently use or are evaluating Superwhisper for sensitive content — or you want to understand the difference between “runs locally” and “keeps all data under your control.”

Is Aqua Voice Safe? Privacy Mode, Training Silence & Verdict (2026)

A current-state safety investigation of Aqua Voice. Walks through Aqua Voice's cloud-only architecture (every dictation request transmits audio to Aqua Voice's servers — no on-device mode), the Privacy Mode mechanics (OFF by default for individuals, with admin-enforceable org-wide enforcement on Teams + Enterprise), and the load-bearing documentation gap: the privacy policy at aquavoice.com/info/privacy effective May 22, 2025 does not address whether stored transcript data is used for AI model training. Covers the SOC 2 Type II attestation through Advantage Partners (Vanta-managed trust center) and what the certification does and does not tell you. Includes a five-question Aqua Voice Safety Decision Tree.

Read this if: You are evaluating Aqua Voice for daily dictation or sensitive content — or you want to understand why a SOC 2 attestation does not by itself answer the AI-training question.

Is Otter.ai Safe? Class Action, Two-Party Consent & Verdict (2026)

A current-state safety investigation of Otter.ai, the leading cloud meeting transcription tool. Walks through Otter's cloud-only architecture (no on-device mode; audio + transcripts stored on Otter servers), the SOC 2 Type 2 attestation and AES-256 encryption baseline, the default-opt-out training pattern (Otter trains on de-identified user data unless you find and flip the setting), and the load-bearing legal story: In re Otter.AI Privacy Litigation, 5:25-cv-06911 (N.D. Cal.) — a consolidated federal class action filed Aug-Sep 2025, consolidated by Judge Eumi K. Lee on October 22, 2025, with a consolidated complaint filed December 5, 2025 and Otter's motion-to-dismiss reply brief filed April 2026. The plaintiffs allege Otter recorded private conversations and trained AI on meeting data without all-participant consent in two-party-consent jurisdictions, citing ECPA, CFAA, CIPA, and two California statutes. Includes a five-question Otter Safety Decision Tree and an Otter Safety Audit checklist.

Read this if: You use OtterPilot for meeting transcription — or you want to understand why the visible-bot-as-implicit-consent model is being litigated and what that means for organizations using meeting bots in mixed-jurisdiction calls.

Is Dragon Safe? Professional, Anywhere, Medical One & Microsoft (2026)

A current-state safety investigation of the Dragon dictation product line, now Microsoft-owned (Nuance acquired March 2022 for $19.7 billion). Walks through the three currently-sold Dragon variants — Dragon Professional v16 ($699.99 Windows-only, mostly on-device), Dragon Anywhere ($14.99/mo or $149.99/yr mobile cloud), and Dragon Medical One ($79–99/user/month on 1–3 year terms, cloud-only on Azure with a signed BAA). Documents the Microsoft acquisition data-perimeter shift to Azure, the full healthcare compliance stack (SOC 2 Type 2, ISO 27001, HITRUST CSF, FedRAMP, HIPAA BAA), the orphaned-Mac-user gap from the 2018 Dragon Mac discontinuation, and Microsoft's March 2025 Dragon Copilot strategy (merged with DAX Copilot). Includes a five-question Dragon Safety Decision Tree and an architectural-alternatives breakdown by user segment (Mac users / healthcare users / legal users / mobile users).

Read this if: You are a Dragon user evaluating which Dragon variant fits your platform and use case — or a Mac user orphaned by the 2018 discontinuation looking for the architectural alternative that Dragon's current line does not cover.

Is Willow Voice Safe? Private Mode, HIPAA & Enterprise Verdict (2026)

A current-state safety investigation of Willow Voice, the YC X25-backed cloud dictation product (Mac + Windows + iPhone + Android). Walks through the privacy-protective Private Mode default — Willow's policy explicitly designates Private Mode as the “(DEFAULT Opt-Out)” for training, the strongest default among major cloud dictation peers we have investigated (Aqua Voice is Privacy Mode off by default, Superwhisper is local-recording on by default, Otter is training opt-out, Wispr Flow is Privacy Mode off by default for individuals). Documents three structural caveats: (1) cloud-first by default — audio still routes through Willow's servers for transcription in both modes; (2) the optional Offline Mode on Mac and iOS is not addressed in the privacy policy effective April 30, 2025; (3) HIPAA is advertised on the homepage and pricing page but the privacy policy text mentions only SOC 2 and GDPR, leaving BAA scope undocumented publicly. Includes a five-question Willow Voice Safety Decision Tree and a Willow Voice Safety Audit checklist.

Read this if: You currently use or are evaluating Willow Voice, especially for sensitive or regulated work — or you want to understand why the most privacy-protective default in the cloud dictation category still leaves documentation gaps that matter for healthcare procurement.

Is Claude Code Safe? Pro/Max vs API Privacy, Aug 2025 Terms Verdict (2026)

A current-state safety investigation of Anthropic's Claude Code that directly addresses the developer confusion around the August 28, 2025 consumer terms update. Walks through the two-tier framework: Consumer (Free, Pro, Max) accounts where Anthropic CAN train on Claude Code prompts and outputs by default — opt out at claude.ai/settings/data-privacy-controls — versus Commercial Terms (Anthropic API, Amazon Bedrock, Google Cloud Vertex AI, Microsoft Foundry, Claude Platform on AWS, Claude for Teams, Claude for Enterprise, Claude Gov) where no training is the documented default. Documents the provider-specific defaults matrix (Bedrock / Vertex / Foundry / AWS all have telemetry / error reporting / /feedback DEFAULT OFF; direct Anthropic API has them on), Zero Data Retention configuration on Claude for Enterprise, the HIPAA BAA path, the local cache at ~/.claude/projects/ (plaintext for 30 days by default), and the environment-variable controls (DISABLE_TELEMETRY, DISABLE_ERROR_REPORTING, DISABLE_FEEDBACK_COMMAND, CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC). Includes a five-question Claude Code Safety Decision Tree.

Read this if: You use Claude Code for any sensitive, regulated, or compliance-audited code — or you want to verify whether your Pro/Max account is currently training Anthropic's models with your code by default after the August 2025 terms update.

AI Tool Privacy Tracker: Verified Reference Matrix for 12 Tools

The cross-product flagship of this cluster. A continuously-updated reference page covering 12 major AI tools across three categories: AI Assistants (ChatGPT, Claude, Gemini, Perplexity), AI Coding Tools (Cursor, GitHub Copilot, Windsurf, Cline), and Voice & Dictation (Voibe, Wispr Flow, Superwhisper, Apple Dictation). Each row is split by plan tier (Consumer / Business-API) — because the same tool typically gives different answers on each side — and every cell links to a primary source (the vendor's own privacy policy, terms, or technical documentation). Includes a Recent Changes timeline of dated policy shifts that move tools between “trains by default” and “does not train.” Reviewed monthly.

Read this if: You want a single answer to “does [AI tool] train on my data?” — or you are choosing between AI assistants, coding tools, or dictation apps and want to weight privacy posture in your decision.

AI and Attorney-Client Privilege After US v. Heppner: What Lawyers Must Know (2026)

In February 2026, Judge Jed S. Rakoff of the Southern District of New York held that a defendant's chats with public Claude were not protected by attorney-client privilege or the work product doctrine. The ruling applied the traditional three-part privilege test and found public AI tools fail every prong: the AI is not an attorney, the privacy policy disclaimed confidentiality, and independent client use lacked counsel's direction. The same third-party-disclosure logic extends to any cloud AI tool that touches privileged content — including dictation, transcription, and meeting-summary tools. This analysis covers the case, the Heppner-Gilbarco split, the public-vs-enterprise-vs-on-device risk spectrum, and a practical post-Heppner checklist.

Read this if: You are a lawyer evaluating AI tools, or anyone curious about how Heppner reshapes privilege analysis for voice and dictation tools.

Accessibility Dictation: A Hub for Hands-Free Voice Typing on Mac

Users with carpal tunnel, RSI, arthritis, post-surgery hands, or ADHD often turn to dictation because typing is the actual cause of the pain or friction. The catch: most dictation apps default to push-to-talk activation, which replaces sustained typing load with sustained held-key load — the same finger-flexion pattern that triggered the original injury. This hub leads with Voibe's Hands-Free Mode (double-tap activation, no key held during speech) as the structural solution, and on-device privacy as the supporting moat — because users dictating about their condition often reference the condition itself in the dictated stream (medications, symptoms, doctor names, insurance codes). Related guides include Best Dictation Software for Carpal Tunnel, How to Type With Carpal Tunnel, Best Dictation Software for Arthritis (joint-protection approach for RA, OA, PsA — including biologic and DMARD vocabulary), Typing With Arthritis (keyboard adaptation and joint protection at work), and Best Dictation Software for Hand Pain (a guide for users with undiagnosed or overlapping conditions).

Read this if: You are evaluating dictation as an ADA accommodation, recovering from hand surgery, managing a chronic condition that limits typing, or work with someone who is.

The Rev.com persona sub-cluster — three guides covering the structurally same problem with different compliance frameworks:

• Rev.com Alternatives for Lawyers and Small Law Firms — anchored on ABA Rule 1.6(c) and the US v. Heppner third-party-disclosure analysis. 8 alternatives including SpeakWrite (1.5¢/word human) and Sonix Enterprise (HIPAA BAA cloud). Pre-calculated 3-year savings on a representative solo workload: 99.2% vs Rev human transcription.
• Rev.com Alternatives for Doctors and Small Practices — anchored on the HIPAA Security Rule and the AI-medical-scribe-vs-transcription category gap. 8 alternatives spanning on-device dictation, AI scribes (Suki, DAX, Heidi), and HIPAA-aligned cloud transcription. 99.4% savings on a 3-doctor 30-min/day workload.
• Rev.com Alternatives for Journalists and Newsrooms — anchored on the state shield-law gap (Branzburg v. Hayes 1972, no federal shield, PRESS Act pending) and third-party-records-holder subpoena exposure. 8 alternatives including Trint Story Builder, Descript transcript-as-AV-editor, and Pinpoint (free Google News Initiative tool). 95.5% savings on a 50-source investigation.

Read these if: You currently use Rev.com for transcription and want to evaluate the dictation, ambient AI, or newsroom-collaboration tools that replace specific portions of that workflow without sending privileged or confidential audio to a third-party processor.

For the complete analysis with pros, cons, and decision guidance, see our offline dictation privacy deep-dive and best offline dictation apps roundup. Healthcare professionals currently using Dragon Medical should also review our Dragon Medical alternatives guide — it covers 7 replacements including on-device tools that keep PHI off cloud servers entirely.

The fastest path to private dictation on Mac: Voibe runs 100% on-device on Apple Silicon, requires no account, and costs $7.50/month, $59/year, or $149 lifetime. Download, install, and dictate — your voice never leaves your Mac.

For a complete walkthrough, see our how to use dictation on Mac guide.

Switching from cloud tools? See our guides to TurboScribe alternatives and SpeakOneAI alternatives for privacy-focused replacements.

Comparing Apple Dictation to the main cloud and open-source alternatives? See Apple Dictation vs Wispr Flow for the upgrade-decision framework around cloud AI dictation, and Apple Dictation vs OpenAI Whisper for the built-in vs open-source model trade-off. For the free-vs-$699 end of the spectrum, see Apple Dictation vs Dragon. Once you have a private dictation tool in place, see our voice input workflow guide for the Talk-Draft-Polish pattern that makes on-device dictation sustainable day-to-day — including a dedicated section on why offline workflows matter for regulated work and private drafting.