I've been wearing two AI recorders. Only one belongs near a client.
Ernie Svenson · email · 2026-07-28 Reader 🖥️
Whether an AI recorder belongs near a client turns not on audio fidelity, price, or even which vendor's security/compliance posture looks stronger on paper, but on one structural property: whether the device only starts capturing after an affirmative, disclosed decision to record (Plaud) or captures everyone in earshot indiscriminately before anyone has consented (Fieldy) — because deleting or never storing the audio afterward cannot retroactively cure a consent failure that already happened at the moment of capture.
practical · practitioner product-trial essay with an embedded ethics/regulatory argument, ending in explicit prescriptive rules — Structured as a hands-on comparison (wearing both devices side by side) that keeps surfacing and discarding candidate criteria — audio quality, data retention, vendor security posture — before landing on the one that actually decides the question: consent-before-capture. The final section converts the analysis directly into 'What I'd recommend' rules, making the practical intent explicit.
Structure
| Part | Role |
|---|---|
| Framing: two devices, doing different jobs | setup |
| The Fieldy's most interesting feature is something it doesn't do (never stores audio; ElevenLabs processes but doesn't retain) | steelman |
| The marketing doesn't match the privacy policy | evidence / vendor-trust comparison |
| The ethics question outranks both devices (NYC Bar Formal Opinion 2025-6, Rule 8.4, all-party-consent states) | main argument |
| Plaud answered a question I'd been asking (physical button on NotePin S; Meta/Amazon acquisitions of always-on wearables) | corroborating market evidence |
| What I'd recommend | operationalization |
| Bottom line | restatement of thesis |
Key terms
- capture vs. storage
Two analytically separate risks: what a device records/hears in the moment (capture) versus what it retains afterward (storage). Svenson's central move is arguing the ethical/consent problem lives entirely in capture and is untouched by how a device handles storage. - client consent (before recording)
An affirmative, disclosed agreement obtained before recording begins, required per bar ethics guidance even where the applicable wiretap statute would technically permit secret recording. - one-party consent / all-party consent (states)
The legal floor set by state wiretap statutes for whether recording without everyone's knowledge is lawful at all — which Svenson treats as necessary but not sufficient, since the ethics rule (Rule 8.4) can demand more than the statute requires. - discoverable
A durable audio file that can be compelled as evidence (subpoenaed) in a fee dispute or malpractice claim — the specific downstream risk of a Plaud-style recording that Fieldy's no-storage design avoids. - Rule 8.4 / deception
The professional-conduct rule Svenson invokes to argue that recording a client without disclosure is misconduct regardless of what the wiretap statute permits — i.e., the ethical duty can exceed the legal minimum. - always-on
A device-design category (Fieldy) that listens continuously and infers conversation boundaries itself, as opposed to a device a user deliberately starts and stops (Plaud) — the design property that, in Svenson's argument, structurally disqualifies a device from client-facing use no matter its data-retention behavior.
Propositions
- explicitcontestedDestroying or never storing the audio does not resolve the ethical/consent problem raised by an always-on recorder: 'the risk isn't what gets stored, it's what gets captured.'conf: highcontradicts ←Can an AI-Powered Scribe Curb Physician Burnout? · p5 — Gallani offers audio deletion as the key privacy safety measure; Svenson's parallel analysis of AI recorders argues deletion doesn't resolve the consent/capture problem — 'the risk isn't what gets stored, it's what gets captured' — since the recording is still processed (and potentially transmitted to a third-party AI vendor) before it's destroyed.
- explicitPer NYC Bar Formal Opinion 2025-6, a lawyer using AI to record, transcribe, or summarize a client conversation must get client consent first, and recording without disclosure is deception under Rule 8.4 even in a one-party-consent state — the ethical duty exceeds the statutory minimum.conf: high
- explicitA vendor's marketing claims ('encrypted end-to-end,' 'HIPAA-compliant') can directly contradict its own privacy policy — which named third-party subprocessors (ElevenLabs, OpenAI) and contained no HIPAA provisions, BAA, or DPA-like terms Svenson could find.conf: high
- implicitcorroboratedNever storing the raw audio is a genuine partial safety argument (no durable, discoverable file) that Svenson found 'more persuasive than I expected,' but he treats it as insufficient once weighed against the consent/capture problem — i.e., data-retention risk and consent risk are different axes, and minimizing one does not address the other.conf: mediumsupports ←This AI notetaker won't sell surveillance to your boss · p6 — Granola's 'never stored audio, and that was the right call' is a second real-world case validating Svenson's observation that never-storing-raw-audio is a genuine, if partial, safety argument (distinct from — and not a full answer to — the retention risk carried by the transcript itself).
- explicitAn always-on wearable indiscriminately captures third parties who never agreed to be recorded (the person at the next table, opposing counsel in a hallway) — a structural property of the device's design, not a fixable engineering defect.conf: high
- implicitIndependently auditable vendor certifications (SOC 2 Type II, ISO 27001, ISO 27701, disclosed/selectable data region) constitute a meaningfully higher trust signal than marketing language claiming encryption and regulatory compliance without a business-associate or data-processing agreement.conf: medium
- implicitcorroboratedOwnership consolidation of always-on personal-recording devices by large tech platforms (Meta acquiring Limitless and discontinuing the Pendant; Amazon acquiring Bee) is itself an additional, separate trust concern for professional use, independent of the device's own privacy engineering.conf: mediumsupports ←AI Sovereignty: Bargaining with Big Tech and the Promise of Full Stack Open Source AI · p8 — Supports the broader pattern that big-tech ownership consolidation of independent AI products is itself a standing risk distinct from any single product's engineering — there Meta/Limitless and Amazon/Bee, here Stripe/OpenRouter and NVIDIA/Hugging Face.
- explicitThe right resolution is context-based tiering rather than a blanket rule: use the consent-gated, start/stop device for anything that could become a record (client meetings, intake, get consent in the first ten seconds), and reserve the always-on device strictly for solo, non-client use — physically removing it, not muting it, before client contact.conf: high
Arguments
The Fieldy (an always-on AI recorder) does not belong near a client, regardless of its data-retention engineering. strength: high
Explicit premises:
- NYC Bar Formal Opinion 2025-6 requires client consent before recording, even in one-party-consent states.
- Recording a client without disclosure is deception under Rule 8.4, regardless of what the wiretap statute permits.
- An always-on device captures everyone in range — including people who never agreed to be recorded — before any consent decision is made.
- Destroying the audio afterward doesn't fix a consent problem that already occurred at capture.
Implicit premises:
- Professional-ethics obligations bind independently of, and can exceed, what a state's recording-consent statute technically permits.
- A device's data-retention protections (e.g., auto-deletion) address confidentiality-after-the-fact but cannot address consent-before-capture.
Inferences:
- Always-on design ⇒ capture precedes any consent decision ⇒ ethical consent requirement cannot be satisfied in client-facing contexts, no matter how the device subsequently handles the data.
Plaud, not Fieldy, is the vendor a professional can actually do diligence on. strength: medium
Explicit premises:
- Plaud publishes SOC 2 Type II, ISO 27001, and ISO 27701, and discloses/lets teams select the data region.
- Fieldy's marketing claims ('end-to-end encrypted,' 'HIPAA-compliant') contradict its own privacy policy, which names third-party processors and contains no HIPAA provisions, BAA, or DPA.
Implicit premises:
- Independently audited certifications are stronger evidence of actual data-handling practice than vendor marketing copy.
- The absence of a BAA/DPA where the marketing implies compliance requiring one is itself informative, not merely inconclusive.
Inferences:
- Marketing claim contradicted by the vendor's own primary source (privacy policy) ⇒ marketing claim is unreliable ⇒ prefer the vendor whose compliance claims are independently verifiable.
Narratives
- Svenson wore both the Plaud and the Fieldy side by side to see which 'earns a permanent spot,' framing the whole piece as a real practitioner trial rather than a spec-sheet comparison — reinforced by his closing note that 'In the Inner Circle we pull tools apart like this before anyone spends money on them — including the ones that don't survive the reading.'
→ Establishes practitioner-experience credibility and signals that the recommendation is earned through use, not vendor marketing or theoretical analysis. - He expected the Fieldy's always-on convenience to pressure Plaud into making its NotePin always-on too; instead Plaud's new NotePin S added a physical button (plus a tap-to-flag feature) and reworked its firmware specifically to cut down on accidental recordings.
→ A small 'the market answered my question' beat that Svenson treats as external corroboration of his own conclusion — the industry moving toward consent-gated design rather than away from it.
Questions addressed
- ✅ Which recorder should a lawyer use for client-facing work?
The Plaud — a consent-gated, start/stop device from a vendor with independently auditable compliance claims. - ✅ Does auto-deleting/never-storing audio solve the ethics and consent problem of an always-on recorder?
No — the consent problem is about what gets captured, not what gets retained afterward. - ✅ Can a buyer trust a device vendor's marketing claims about encryption and regulatory compliance at face value?
No — marketing language can directly contradict the vendor's own privacy policy; the policy (and independent certifications) are the actual evidence, not the homepage copy. - ⭕ Does the same consent-before-capture principle apply with equal force to professionals who aren't lawyers and have no bar-association ethics opinion or comparable professional-conduct rule governing them?
Gaps
- incompleteThe entire ethical argument is anchored to lawyer-specific authority — a bar association ethics opinion and Rule 8.4 — plus state wiretap/all-party-consent statutes. Svenson never addresses whether the same consent-before-capture principle should bind non-lawyers in equally confidentiality-sensitive settings (an engineering leader in a vendor negotiation, an HR conversation, an M&A due-diligence call, a performance discussion), where no professional body issues an ethics opinion and the only floor is the (often much lower) state wiretap statute. His own framework — 'the risk isn't what gets stored, it's what gets captured' — has nothing legal-profession-specific about it and generalizes cleanly, but he doesn't make that move himself.
Read next: Professional-ethics literature from other regulated fields (medical, financial advisory, HR/employment law) on AI recording and consent; general workplace-recording guidance for roles with no governing ethics body but real confidentiality stakes (vendor/client meetings, performance conversations). - incompleteEvery recommendation is framed as an individual practitioner's personal device choice ('take the Fieldy off... not muted, off') rather than as an organizational policy question. There's no discussion of how a firm, company, or AI-governance function would mandate, audit, or enforce a consent-gated-vs-always-on distinction across an entire client-facing staff — the exact kind of proportionate, risk-tiered control that would need to be built into onboarding, IT provisioning, or a written AI-use policy rather than left to individual judgment and memory.
Read next: Enterprise AI-usage policy design; how organizations operationalize personal-device recording rules (BYOD security policy analogs) for client-facing roles. - incompleteSvenson's safety argument for Fieldy rests on 'it never keeps the audio' — but both devices produce a transcript and summary, and he never asks whether that transcript (stored in the Fieldy or Plaud app, presumably indefinitely) carries the same discoverability and confidentiality exposure as the original audio would have. If the text output persists, 'no durable file' may be true only of the audio layer, and the actual safety delta between the two devices could be much smaller than the piece implies for the one axis (discoverability) where he grants Fieldy real credit.
Read next: Fieldy's and Plaud's actual data-retention terms for transcripts/summaries specifically (as opposed to raw audio); case law on whether AI-generated transcripts of privileged conversations are treated as work product or as a discoverable record in their own right.closed byThis AI notetaker won't sell surveillance to your boss — Pedregal confirms Granola treats the transcript — not the audio — as the actually-retained, valuable artifact: audio was never stored, but verbatim transcripts persist by default and are only now being researched for auto-deletion/compression. This is real-world confirmation of Svenson's concern that 'no audio storage' does not mean 'no durable, discoverable record' — the transcript is where the retention and discoverability risk actually lives, exactly as g3 hypothesized about a different device.
Related documents
- AI Sovereignty: Bargaining with Big Tech and the Promise of Full Stack Open Source AI
- How the Escalating AI Wars Benefit You
- How governments and organizations are leveraging Google's AI breakthroughs for crisis resilience
- How Gemini Flash agents are helping a Michigan dairy farmer
- Expanding Daybreak as the Cyber Defense Window Narrows | OpenAI
- The Tokens You Can't Wait For
- Responsible AI Is Becoming a Growth Strategy