Feature Data Privacy
Neural Data Becomes Sensitive Data in Delaware on January 1. Run the Signal Test First.
Delaware HB 380 makes neural data sensitive data on Jan 1, 2027, joining California, Colorado, Connecticut and Montana. Here is the measured-vs-inferred test and the vendor evidence.
Table of Contents
TL;DR
- Delaware HB 380 adds neural data to the sensitive-data definition at 6 Del. C. § 12D-104, effective January 1, 2027, behind California, Colorado, Connecticut and Montana.
- The statutory line is measurement of central nervous system activity, not the word “neurotech” on a vendor deck. A heart rate variability reading is not neural data; an EEG signal from the same earbud is.
- Sensitive data means consent, a necessity-and-proportionality argument, a documented data protection assessment, and a five-way conversation with the vendor about firmware, subprocessors, and what happens when the enrollment template changes.
A vendor demo lands in your inbox: earbuds that read attention during a budgeting flow, pitched as “step-up authentication without the friction.” Legal asks the only question that matters — is this PHI, is it biometric, is it something new — and the deck has an architecture diagram instead of an answer.
By January 1, 2027 the answer in Delaware is statutory. HB 380 puts neural data inside the sensitive-data definition, and sensitive data comes with a consent obligation, a proportionality argument, and a documented assessment.
What the statute actually says
The amendment rewrites item (30) of the definitions in 6 Del. C. § 12D-104. The operative line: sensitive data includes “Neural data that is generated by measuring the activity of an individual’s central nervous system.”
Read the mechanics underneath it. Section 12D-106(a)(4) bars processing sensitive data unless the consumer consents and the processing is “reasonably necessary and proportionate to the disclosed purposes.” Section 12D-106(a)(6) requires a revocation path at least as easy as the consent path, with processing to stop within 15 days of revocation. Section 12D-106(b)(12) requires consent obtained through “a clear and conspicuous notice of the sale of sensitive data” before any sale. Section 12D-108(a)(1)(d) makes processing of sensitive data a trigger for a documented data protection assessment. Enforcement runs solely through the Delaware Department of Justice.
Delaware is late to this, not early. SB 1223 (Chapter 887, Statutes of 2024) added neural data to sensitive personal information under the CCPA. HB 24-1058, effective August 7, 2024, folded neural properties into “biological data” under the Colorado Privacy Act with opt-in consent. Connecticut’s Public Act 25-113 took effect July 1, 2026, covering central nervous system data only. Montana’s SB 163 extends its Genetic Information Privacy Act to “neurotechnology data.”
Which means the real work is not reading Delaware. It is deciding, signal by signal, whether what your product or vendor touches is covered anywhere you operate — and writing down how you decided.
The signal test
Run this before procurement runs the demo. Three questions, in order, and the answers are the record.
| Question | Covered | Not covered | Why it lands that way |
|---|---|---|---|
| What is physically measured? | Electrical activity at the scalp or cortex (EEG, ECoG, implants) | Pulse, HRV, skin conductance, skin temperature, pupil diameter | Statutes reach central nervous system measurement; the autonomic signals are downstream |
| Is the measurement direct or inferred? | Raw and near-raw signal from the sensor | A “stress score” computed from a camera-based facial read | California excludes data inferred from non-neural information; a selfie liveness check is biometric, not neural |
| Does the vendor ship a derived feature instead of the signal? | Feature computed from neural input, retained or transmitted | Feature the vendor swears never leaves the device | Delaware also captures inferences revealing a listed category, so the derived layer needs its own call |
Two of these produce the same mistake in opposite directions. Teams assume anything brain-adjacent is covered and build consent for a heart rate strap, which wastes effort and teaches the product team that privacy review is theater. Teams also assume a “focus score” is a clean exemption because it is not raw EEG, which is an argument, not a conclusion.
Write the determination down with the vendor’s data flow diagram attached. “Not neural data on the face of § 12D-104” is a defensible line for a chest strap. “Derived score, treated as in scope pending vendor confirmation of feature provenance” is a defensible line for the earbuds.
Where a bank or fintech actually meets this
The GLBA reflex arrives fast: Delaware exempts financial institutions and affiliates subject to GLBA Title V, so the deposit and lending data side is largely out. The exemption is drafted at the data level in this amendment, not the entity level, and four seams stay open.
Processors and non-bank counterparties. A BaaS program’s fintech partner is not a GLBA financial institution. Neither is the analytics vendor scoring attention. Their processing is in scope independent of the bank’s status.
Non-GLBA product surfaces. A wellness rewards program, a financial-wellness coaching feature, a consumer research panel — none of it is GLBA data, all of it can carry a neural signal.
Contract flow-down. Sponsor banks and enterprise customers are already writing privacy addenda that reference state sensitive-data standards. The applicability and contract analysis is where those clauses get negotiated; if the addendum says “at least as protective as the DPDPA,” the neural line binds whether or not Delaware enforces against you directly.
The consumer-facing claim. “We read your focus” is a marketing statement with a data classification behind it. If the classification says sensitive, the privacy notice and the consent screen have to match.
The evidence pack for an EEG vendor
A neurotech vendor questionnaire is not a privacy questionnaire with extra rows. It needs the things that only exist in this category. Ten items, and the onboarding gate does not open until each is answered in writing.
- Signal provenance attestation — which sensor, which physiological source, sampled how, and whether the output is raw, filtered, or a derived feature.
- Named legal theory for the vendor’s own classification, with the specific statute text they are reading.
- Version registry — firmware version, model version, and the date the enrollment template was generated, per user.
- Enrollment and re-enrollment mechanics — what happens to a stored template on a firmware update or model retrain, and who re-consents.
- Subprocessor list specific to the neural pipeline, including any cloud inference provider.
- Retention and deletion by layer: raw signal, features, templates, and anything folded into model weights.
- Training rights — whether the vendor trains on your users’ signals, and whether that survives contract termination.
- Consent surface screenshots, with the disclosure text version identifier, captured at onboarding.
- Revocation path — the actual UI, plus the measured time to stop processing, tested against the 15-day limit.
- Assessment support — the vendor artifact your data protection assessment can cite, and notice of material pipeline changes.
Items 3, 4, and 7 are where the deck and the contract usually disagree. Attention models drift; a template enrolled on firmware 1.4 and matched by a model retrained in March is a different biometric comparison than the one the user consented to. Version-pin it and require notice on change, the same way you would for any material vendor change.
Consent that survives an audit
Delaware’s standard is consent plus necessity plus proportionality, and a consent capture that only records a checkbox fails all three tests of usefulness.
Build the record so a stranger can reconstruct the deal: identifier, timestamp, disclosure text version, purpose enumerated at the category level (“step-up authentication for card-not-present transactions” rather than “security”), the specific data elements named, and the vendor or subprocessor recipients. Then write the necessity sentence into the assessment — why the neural signal beats the alternative for this purpose. For step-up auth, the argument is plausible: a liveness-plus-attention check is harder to spoof than a static selfie. For a budgeting app adding EEG to a flow that already has transaction data, the argument is weaker and should say so.
Proportionality is where product pushback lands hardest, and it usually arrives as “can we keep the raw signal.” The workable answer is a retention ladder: raw for a defined window to support retraining and dispute resolution, features for the product lifetime, templates as long as the account is open. Every rung needs a stated purpose.
If the neural data is sold or shared for value, § 12D-106(b)(12) raises the bar to clear and conspicuous sale notice, and the sale inventory entry needs the recipient category and purpose. Biometric privacy statutes stack on top rather than replacing this — the BIPA and vendor exemption analysis is the companion piece, because a headset-derived template can be argued into biometric definitions in Illinois and Texas.
The 10-week sequence before January 1
Delaware’s date is fixed and the assessment cannot be written in December.
Weeks 1–2. Inventory every place a nervous-system signal touches the business: shipped features, pilots, vendor roadmaps, and the two decks sitting in procurement with no privacy review.
Weeks 3–4. Run the signal test on each and record the determination with the evidence. Route anything ambiguous to counsel instead of resolving it in a product channel.
Weeks 5–6. Draft or amend the data protection assessments for confirmed in-scope processing. Consent architecture, retention ladder, and revocation mechanics get designed here, with engineering in the room.
Weeks 7–8. Paper the vendors. Add the neural-data schedule, the version-registry obligation, the change-notice clause, and the assessment support term.
Weeks 9–10. Update the privacy notice, ship the consent screens, and tabletop a revocation request to time the 15-day clock before a consumer starts it.
If the pipeline is empty, the sequence shortens to a scoping memo and a standing clause in the vendor template. That is a legitimate outcome. Discovering an in-scope pilot in February is not.
So What?
Neural data is the rare privacy category where the classification question is genuinely unsettled at the edges and genuinely settled at the center. Raw central nervous system measurement is covered in five states; a heart rate reading is not. The gap between those two facts is where programs get sloppy, because both arrive in the same earbud.
Decide per signal, write down the reasoning, and make the vendor prove the pipeline. The statute asks for a necessity argument, not a vibe.
The Data Privacy Compliance Kit has the assessment scaffolding, consent record fields, and vendor schedules to run this without starting from a blank page.
FAQ
Does the 15-day revocation clock start when the consumer submits or when we confirm? On receipt of the request, per § 12D-106(a)(6). Do not wait for identity verification to start the timer, but do document verification separately.
We only ship a derived “cognitive load” score. Are we out of scope? Probably in scope, and the burden is on you to show otherwise. Get the vendor’s feature provenance in writing and record the call either way.
Does an existing HIPAA business associate agreement cover this? Only for the covered-entity workflow. A consumer wellness feature outside a provider relationship is not HIPAA data, and the state analysis applies on its own.
What if the vendor refuses the provenance attestation? Treat it as a diligence gap, not a deal-breaker, and reduce scope: features only, no raw retention, shorter term, tighter termination. Refusal is informative — it usually means they have not classified their own pipeline.
◆ Need the working template?
Start with the source guide.
These answer-first guides summarize the required fields, evidence, and implementation steps behind the templates practitioners search for.
◆ Related template
Data Privacy Compliance Kit
Which of the 23 state privacy laws apply to your fintech after GLBA, plus the GLBA checklist, request tracker, assessments and vendor terms to comply.
◆ Immaterial Findings · Weekly
Sharp risk & compliance insights. No fluff.
◆ FAQ
Frequently asked questions.
Is neural data sensitive data under Delaware law?
Does an EEG headband used for login actually count, or only medical brain implants?
What about a focus or attention score computed from the raw signal?
We are a bank, so does the GLBA exemption cover us?
Author
Rebecca Leung
Rebecca Leung has 8+ years of risk and compliance experience across first and second line roles at commercial banks, asset managers, and fintechs. Former management consultant advising financial institutions on risk strategy. Founder of RiskTemplates.
◆ Related framework
Data Privacy Compliance Kit
Which of the 23 state privacy laws apply to your fintech after GLBA, plus the GLBA checklist, request tracker, assessments and vendor terms to comply.
◆ Keep reading
Related posts.
Data Privacy
Consent Receipts for Sensitive-Data Sales Must Survive Five Years. Most Records Are Two Fields.
Delaware HB 380 requires a five-year consent record produced with the DPIA. Here is the field set, the versioning rule, and the deletion conflict.
Nov 16, 2026
Data Privacy
Delaware Just Narrowed the GLBA Privacy Exemption. Run the Applicability Test Before January 1.
Delaware HB 380 takes effect January 1, 2027, cuts the threshold to 10,000 consumers, and strips the entity-level GLBA exemption from fintechs. Scope it now.
Nov 14, 2026
Data Privacy
Texas's AI Law Changed What Your AI Vendors Owe You. TRAIGA's TDPSA Amendments Have Been in Effect Since January.
The Texas Responsible AI Governance Act (TRAIGA, HB 149) took effect January 1, 2026, amending the TDPSA to add AI system processor obligations. Federally insured financial institutions get a carve-out from TRAIGA's non-discrimination rules — but not from the processor amendment. Here's what fintechs and banks using AI for Texas customers need to assess now.
Oct 11, 2026