Does a BAA with a cloud AI scribe cover the risk of a DBT chain-analysis session audio being subpoenaed?

No. A Business Associate Agreement establishes the vendor's HIPAA obligations — permitted uses, safeguards, breach notification — but it does not constitute a privilege or basis for withholding records in legal proceedings. A subpoena issued to the cloud AI scribe vendor for session audio is addressed to the vendor, not to the therapist. The vendor's obligation in response to a valid subpoena is governed by the applicable law, not by the BAA with the therapist. For DBT clients, who disproportionately make disclosures that become relevant in custody disputes, protective order proceedings, and civil litigation, this is a material risk that a BAA does not resolve.

Does 42 CFR Part 2 apply to DBT chain-analysis notes when substance use appears as a vulnerability factor?

42 CFR Part 2 applies to records of substance use disorder treatment programs — specifically federally assisted programs whose primary purpose is substance use disorder treatment. A standard outpatient DBT practice is almost certainly not a covered Part 2 program, even if substance use appears in chain-analysis vulnerability factors or precipitating events. However, DBT is increasingly practiced in integrated behavioral health settings that also provide substance use disorder treatment, and if the setting qualifies as a Part 2 program, the chain-analysis notes documenting substance use could be subject to Part 2's more restrictive disclosure requirements. The safest posture in any integrated or dual-diagnosis setting is to consult with a health privacy attorney before using any AI scribe — cloud or on-device — for sessions that may contain substance use disclosures.

Blog · Clinical Workflows

DBT chain-analysis notes and AI scribes: when the most sensitive session content meets the cloud

Q: Are DBT chain-analysis notes different from standard SOAP notes for HIPAA purposes?

Both are progress notes under HIPAA — they are part of the designated record set and subject to the same access and disclosure rules. The distinction that matters for AI-scribe risk is content, not category: a chain-analysis note documents specific behavioral detail (self-harm methods, precipitating events, named third parties, step-by-step crisis narratives) that standard SOAP notes typically abstract or omit. When a cloud AI scribe processes the session audio to draft that note, it receives the full raw disclosure — not just the final clinical summary. The behavioral specificity that makes chain analysis clinically valuable is the same specificity that makes the session audio significantly more sensitive than a standard session recording.

Q: How does the DBT consultation team model interact with AI-scribe data custody?

DBT requires therapists to participate in a weekly consultation team — a group of DBT-trained clinicians who review cases and support skill adherence. When chain-analysis notes drafted by a cloud AI scribe are shared with a consultation team (even in redacted form), the data has moved from the cloud scribe's servers through the therapist's EHR into a second set of covered entities with their own data-handling practices. This extends the data custody chain beyond what the scribe's BAA covers. With on-device note drafting, the note originates on the therapist's device. Third-party names and sensitive behavioral details can be redacted locally before sharing with the consultation team — without any coordination request to a cloud vendor about data already held on their servers.

2026-05-31 · 2,100 words · All posts

TL;DR

DBT chain-analysis notes document behavioral specificity — self-harm methods, precipitating events, crisis narratives, named third parties — that standard SOAP or DAP notes abstract or omit.
When a cloud AI scribe processes a DBT session, it receives the full audio containing every disclosure. The note it returns is a clinical summary; the audio it received is much more.
The DBT consultation team model extends the data-custody chain: notes drafted by a cloud scribe may be shared with peers, each in their own system, before the scribe's copy has been accounted for.
A BAA does not prevent a subpoena. DBT clients — who disproportionately disclose precipitating events that appear in custody, civil, and criminal proceedings — face material legal risk that a signed agreement cannot resolve.
On-device processing means the chain-analysis audio never leaves the therapist's Mac. There is no cloud copy to subpoena, no vendor retention window to audit, and no third-party server holding a record of what the client disclosed.

Dialectical Behavior Therapy is the gold standard for treating borderline personality disorder and chronic self-harm. It is also, by design, one of the most documentation-intensive therapy modalities in clinical practice. The centerpiece of DBT documentation is the behavioral chain analysis: a structured, step-by-step narrative of a problem behavior that begins with a precipitating event, traces the sequence of emotional, cognitive, and behavioral links that led to the behavior, and concludes with a solution analysis identifying intervention points for next time.

Chain-analysis notes are not just different in format from a standard SOAP or DAP note. They are categorically more sensitive. A SOAP note documents the therapist's clinical observations and assessment. A chain-analysis note documents the client's own narration of a crisis — specific disclosures about self-harm behaviors, precipitating traumas, named third parties, and the step-by-step internal experience of a clinical emergency. When that session audio is processed by a cloud AI scribe, the vendor receives far more than a routine clinical encounter. This post examines what cloud scribes actually receive from a DBT chain-analysis session, why the standard BAA framework does not resolve the resulting exposure, and why on-device processing eliminates a category of risk that no contractual instrument can.

What a DBT chain-analysis note actually contains

DBT was developed by Marsha Linehan in the late 1980s as a treatment for chronically suicidal patients with borderline personality disorder. The behavioral chain analysis is both a therapy technique and a documentation form. A completed chain-analysis note documents a structured sequence:

Precipitating event — the specific external event that initiated the chain. This is not an abstraction ("interpersonal conflict") but a named, dated, specific event: a particular argument with a named person, a specific rejection message, a job loss, an anniversary of a trauma.
Vulnerability factors — background conditions that lowered the client's distress threshold at the time of the precipitating event: disrupted sleep, missed medication, recent substance use, physical illness, isolation. These are documented specifically because they are intervention targets.
Chain of links — the sequential cognitive, emotional, and behavioral steps from the precipitating event to the problem behavior. A detailed chain analysis may document 8 to 20 discrete steps, each capturing the client's internal state and overt behavior at that point in the sequence.
Problem behavior — the specific target behavior documented with clinical precision. For a self-harm chain, this includes method, body location, duration, and severity. For a suicidal behavior chain, it includes ideation content, plan specificity, means access, and intent. This level of detail is clinically necessary to track patterns and calibrate interventions over time.
Consequences — short- and long-term effects of the behavior on the client's life, relationships, physical health, and treatment trajectory.
Solution analysis — an identification of which links in the chain could have been addressed differently, and what skills or interventions would have altered the outcome at each point.

A chain-analysis note for a self-harm session is a dense clinical document — typically several hundred to more than a thousand words — that captures the client's lived experience of a crisis with behavioral specificity that would be inappropriate in most other therapeutic contexts. It names people. It describes methods. It traces the internal narrative of a mental health emergency in the client's own frame.

Why chain-analysis content is categorically more sensitive

Standard progress notes — SOAP, DAP, BIRP, GIRP — document clinical observations, the therapist's assessment, and the treatment plan. They describe what the therapist observed and concluded. Content that is not clinically necessary for the record typically is not included.

Chain-analysis notes work differently. Their clinical value is precisely their specificity. Consider what is typically absent from a standard progress note but present in a chain analysis:

Self-harm method documentation — location, instrument, severity — because tracking patterns across sessions is how DBT clinicians measure progress on Target 1 behaviors.
Suicidal ideation with plan detail — means specificity and access are assessed and documented because lethality assessment requires it. A chain analysis on a suicidal behavior chain typically contains more specific ideation content than anything in the client's psychiatric intake.
Precipitating event narratives that name specific third parties. An argument with a romantic partner, a family member's behavior during a holiday visit, a supervisor's action at work — these individuals are named because the precipitating event cannot be clinically analyzed in the abstract. The same individuals may be parties to custody disputes, protective orders, or litigation in which the client is involved.
Substance use documented as a vulnerability factor. If the client used alcohol or cannabis before the crisis, that use appears in the vulnerability factors section. This creates potential 42 CFR Part 2 questions in integrated behavioral health settings.
The client's own reported internal monologue during the chain — thoughts, emotional states, urges — often in close paraphrase, because DBT's behavioral analysis requires reconstructing the cognitive sequence that led to the behavior.

None of this content appears because the therapist failed to be appropriately brief. It appears because it is clinically necessary. But the same content that makes a chain-analysis note clinically valuable makes the session audio that generated it significantly more sensitive than a standard therapy recording.

The DBT consultation team model and data custody

A frequently overlooked dimension of DBT documentation is structural: DBT, as an evidence-based treatment, requires therapist participation in a weekly consultation team. The consultation team is not optional for standard DBT implementation — it is part of what distinguishes DBT from an eclectic use of DBT skills. Marsha Linehan's original protocol and the work of the DBT-Linehan Board of Certification both treat consultation team participation as a treatment adherence requirement.

The consultation team reviews cases, including chain-analysis notes. A therapist practicing in a two-person private practice, a group practice, or a solo practice who participates in a community-based consultation team shares chain-analysis content beyond their own records. This extends the data custody chain in ways that the cloud scribe's BAA does not cover.

When a chain-analysis note is drafted by a cloud AI scribe, the custody sequence looks like this: the vendor's servers receive and hold the audio → the vendor returns a draft → the therapist edits and pastes the note into their EHR → the EHR holds the note → the therapist shares it (or a version of it) with a consultation team of peers, each in their own EHR system. The vendor's BAA covers step one. It does not cover what happens downstream in the consultation team's review process, because the BAA is a bilateral agreement between the vendor and the therapist.

The practical question this raises: if the vendor's retention window for session audio runs for 30 or 90 days, and the therapist wants to redact or minimize a chain-analysis note before sharing it with a consultation peer, they cannot retroactively limit what the vendor already received. The audio is on the vendor's servers from the moment of upload, for whatever period their privacy policy specifies.

What the cloud AI scribe receives from a DBT chain-analysis session

The output the cloud AI scribe delivers is a structured chain-analysis note. That output is exactly what the therapist needs. But the input the vendor received to produce that output is the full session audio — typically 45 to 55 minutes of recorded session content.

For a DBT chain-analysis session, that audio includes the client's detailed account of the problem behavior chain: the precipitating event narrative (naming specific people and circumstances), the step-by-step internal experience of the crisis, and the behavioral specifics of whatever Target 1 behavior the chain examined. The cloud scribe's speech-to-text model converts the audio to a transcript — a complete, verbatim text record of the session — before the language model processes the transcript into a structured note. Both the audio and the intermediate transcript are held by the vendor for whatever period their privacy policy specifies.

A category-by-category breakdown of what cloud scribes receive and retain is covered in the data-flow explainer. For DBT specifically, the categories that differ from a standard session are: self-harm behavioral detail (method, severity), precipitating event narratives (named third parties), suicidal ideation content (plan specificity where assessed), and vulnerability factor documentation (substance use where present). Each of these categories carries risks in potential legal proceedings that the structured note summary does not convey — because the summary abstracts and condenses what the audio captured in full.

Duty to warn, mandated reporting, and the note as legal evidence

DBT clients make safety disclosures. This is an expected feature of DBT treatment with high-risk populations, not a clinical exception. The chain-analysis protocol specifically brings Target 1 behaviors (suicidal and self-harm behaviors) and Target 2 behaviors (therapy-interfering behaviors) into explicit focus. A therapist practicing standard DBT will regularly document sessions where Tarasoff-adjacent duty-to-warn assessments were conducted, where mandated reporting obligations were evaluated, and where the client's disclosures had legal relevance beyond the clinical record.

As the subpoena risk explainer details, a subpoena can reach a cloud AI scribe vendor directly for session audio. This is not a theoretical risk for DBT practices. Consider the scenarios where chain-analysis content typically becomes legally relevant:

Custody proceedings — when the precipitating event named in a chain analysis involves a co-parent, and the chain documents the client's behavior during that conflict (including any self-harm or suicidal behavior that occurred during the co-parenting relationship), the other party's attorney may subpoena the chain-analysis notes as evidence of parental fitness or mental health status.
Civil litigation — when a precipitating event involves a third party (an employer, a school, a landlord) who is later sued, the chain-analysis note naming that party as a precipitating event may be discoverable in the litigation.
Protective order proceedings — when a named person in a chain-analysis precipitating event is the subject of a protective order application or a restraining order, the chain-analysis note documenting their alleged behavior may be sought as evidence.
Criminal proceedings — when a precipitating event involves a crime (assault, domestic violence, sexual abuse), the chain-analysis note may be sought by prosecutors or defense counsel.

In each of these scenarios, a subpoena to the cloud AI scribe vendor can reach the session audio independently of the therapist's records. The vendor holds the audio; the subpoena goes to the vendor. The therapist's privilege assertion, if any, would need to be raised in the jurisdiction of the subpoena — which may be a state the therapist has never practiced in. The BAA does not constitute a privilege or a bar to valid legal process.

The on-device argument for DBT documentation

When a therapist uses TherapyDraft for DBT chain-analysis notes, the session audio is processed entirely on the therapist's Mac using Whisper.cpp for transcription and a quantized local language model for note drafting. The audio does not leave the device. The intermediate transcript does not leave the device. The chain-analysis draft is generated on the machine in the therapist's office and stays there until the therapist pastes it into the EHR.

This changes the risk calculus for DBT specifically because of the content characteristics examined above:

No vendor copy of the self-harm disclosure. There is no cloud server holding the audio of the client describing the method, location, and sequence of a self-harm episode. A subpoena cannot reach a vendor that holds no records.
No retention window to audit. Cloud scribes retain audio for 30 to 90 days (or longer) per their privacy policy. On-device processing produces no audio copy outside the therapist's control — there is nothing to ask the vendor to delete because nothing was ever transmitted.
Third parties named in precipitating events are not in any external database. The client's account of the argument, the name of the person involved, the date and circumstances of the event — none of this leaves the device. The chain-analysis note that reaches the EHR was generated without any of those specifics ever touching a cloud vendor's infrastructure.
Consultation team sharing is within the therapist's control. Because the note was generated on the therapist's device, they can decide exactly what level of detail to share with consultation peers before anything leaves their machine. No retroactive data-destruction request to a vendor is needed — the data never existed anywhere but on the therapist's Mac.

This is not a complete resolution of all DBT documentation risk — the EHR where the note is pasted holds a copy, the telehealth platform holds the video recording if the session was remote, and the consultation team receives whatever the therapist chooses to share. But the AI-scribe layer is removed from the risk equation entirely. The vendor who processes the chain-analysis session audio is TherapyDraft running on the therapist's own hardware, which means the vendor is the therapist themselves.

Practical workflow: DBT chain analysis with on-device note drafting

TherapyDraft handles DBT chain-analysis format natively through one-shot template matching. The therapist provides five examples of their own completed chain-analysis notes — notes from their EHR, redacted or from training cases — and the local language model learns to structure output in that exact format for future sessions.

The session-to-note workflow:

Record the session. Use TherapyDraft's built-in recording (Mac microphone) or drag in a saved audio file if you record separately. The file stays on your device from the first moment.
Select the chain-analysis template. Choose the DBT chain-analysis format from your saved templates, or select a one-shot example from your own prior notes. The local model uses this to structure the output.
Transcription and draft generation. Whisper.cpp runs on Apple Silicon — a 50-minute session transcribes in under 5 minutes on an M2 or later. The local model then drafts the structured chain-analysis note: precipitating event, vulnerability factors, chain links, problem behavior, consequences, solution analysis.
Review and edit. The draft appears in the app. You review, adjust clinical language, edit any links in the chain where the model missed nuance, and finalize the solution analysis. This step takes the same 2 to 3 minutes it would take to edit any AI-assisted note draft.
Paste to EHR. Copy the finalized note to clipboard. Paste into SimplePractice, TheraNest, TherapyNotes, or whichever system you use. The note is in the chart; the audio and intermediate transcript stay on your Mac, subject to your own retention decisions — not a vendor's policy.

For consultation team review: because the draft was generated and finalized on your device, you make redaction decisions before anything leaves your machine. You can remove identifying details from the precipitating event narrative, abstract third-party names to initials or roles, and share a peer-ready version with your team — without any of the original audio, transcript, or unredacted draft ever having resided on a third-party server.

DBT chain-analysis notes that never touch a cloud server.

TherapyDraft processes every DBT session entirely on your Apple Silicon Mac. Self-harm disclosures, precipitating events, named third parties — none of it leaves your device. No vendor retention window. No subpoena surface outside the EHR.

Join the waitlist — 10 free sessions, no card

Frequently asked questions

Are DBT chain-analysis notes different from standard SOAP notes for HIPAA purposes?

Both are progress notes under HIPAA — part of the designated record set, subject to the same access and disclosure rules. The distinction that matters for AI-scribe risk is content, not category. A chain-analysis note documents behavioral specificity — self-harm methods, precipitating events named in detail, third-party names, crisis narratives — that standard SOAP notes typically abstract or omit. When a cloud AI scribe processes the session audio to draft that note, it receives the full raw disclosure. The clinical value of chain analysis is precisely its specificity; that same specificity makes the underlying session audio significantly more sensitive than a routine clinical recording.

Does a BAA with a cloud AI scribe protect against a subpoena for DBT session audio?

No. A Business Associate Agreement establishes the vendor's HIPAA obligations — permitted uses, safeguards, breach notification timelines — but it does not constitute a privilege or a basis for withholding records in legal proceedings. A valid subpoena issued to the cloud scribe vendor for session audio is addressed to the vendor, not to the therapist. The vendor's response is governed by applicable law. For DBT clients who make disclosures that become relevant in custody, civil, or criminal proceedings — which is a predictable feature of high-risk populations, not a rare edge case — this is a material exposure a signed agreement does not resolve.

Can TherapyDraft handle DBT chain-analysis format, or only SOAP and DAP?

TherapyDraft supports DBT chain-analysis format through one-shot template matching: you provide five examples of your own completed chain-analysis notes, and the local model structures future drafts in that exact format — precipitating event, vulnerability factors, chain links, problem behavior, consequences, solution analysis. Because the model runs entirely on your Mac, the calibration examples never leave your device. Cloud AI scribes offer similar template matching but require your example notes to be uploaded to and processed on their servers.

How does the DBT consultation team model affect AI-scribe data custody?

DBT requires therapist participation in a weekly consultation team that reviews cases and chain-analysis notes. When notes drafted by a cloud AI scribe are shared with a consultation team — even in redacted form — the data has moved from the vendor's servers through the therapist's EHR to a second set of covered entities with their own data-handling practices. The vendor's BAA covers only the first step in this chain. With on-device drafting, the note originates on your device. You decide what level of detail to share with consultation peers before anything leaves your machine, and there is no retroactive request to send a cloud vendor about audio they already hold on their servers.

Does 42 CFR Part 2 apply if substance use appears in a DBT chain analysis as a vulnerability factor?

42 CFR Part 2 applies to records of federally assisted substance use disorder treatment programs — not to all clinical records that mention substance use. A standard outpatient DBT practice is almost certainly not a covered Part 2 program, even if substance use appears in chain-analysis vulnerability factors. However, DBT is increasingly delivered in integrated behavioral health settings that do qualify as Part 2 programs. In those settings, chain-analysis notes documenting substance use as a vulnerability factor or precipitating event could be subject to Part 2's more restrictive disclosure requirements. If your practice has any integrated substance use treatment component, consult a health privacy attorney before using any AI scribe for sessions that may contain substance use disclosures.