Skip to content

Glossary of worker-built app terms

This site uses one canonical term for each idea in its body copy, and maps the common search terms ("vibe coding", "vibe-coded", "citizen developer", "citizen-developed", "shadow AI") to those terms. Each entry gives a short definition and links to the page that owns the topic. If an entry and its canonical page ever differ, the canonical page wins.

Who this is for

Everyone: builders, app owners, IT, security, legal and executives. Look up a term, or link to a definition from your own policies. Reading time: dip in as needed.

Terms marked search term appear on this site only in titles, descriptions and opening paragraphs; body copy uses the canonical term. Legal terms are in their own section, because their meaning depends on the jurisdiction.

Terms A to Z

A

AI app builder
A tool that generates and hosts an app from natural-language prompts, often with a database and sign-in. Not "vibe coding platform". See low-code vs. AI app builders.
AI-assisted building
Using AI tools to generate or change an app's code or configuration, by anyone. The site's body-copy term for what searchers call vibe coding.
AI feature
Something an app does by calling an AI model at runtime, such as drafting replies. It's a different risk from AI-assisted building and needs its own controls. A tool's built-in AI features (an AI formula in a spreadsheet, an AI step in a workflow tool) count when they process the app's data at runtime; one that's on the approved-tools list for the data class already meets the M1 approval gate. An app with an AI feature can be an AI system in law; AI-written code isn't one. See AI feature security.
AI gateway
A model-access layer between apps and AI providers that enforces approved models, logs use and caps spend. As of October 2026 we found no independent data on how widely it's used. See AI spend.
App owner
The accountable person for an app once it's in use, recorded as a role with a named successor. May differ from the builder. See app ownership.
App registry
The record of every worker-built app: owner, data, risk tier, support level, status and retire-by date. Not "inventory" or "CMDB". See the app registry and the app registry template.
Audit-scope flag
An intake question, not a tier: does the app help deliver a service covered by a SOC 2 or similar attestation, produce evidence that a control is operating, hold privileged access, or touch financial reporting? If yes, it's routed to the compliance owner. See the audit-scope flag.

B

Baseline
A measurement of a process before a change (cycle time, errors, rework), so an app's effect can be measured, not guessed. See baselining.
Builder
The person who builds a worker-built app: a non-developer or a professional developer. Pages say which they mean. See the builder handbook.
Business review
At Important and above, a review by any colleague who knows the process, covering the logic, the data used and whether access matches the intent. See business review and technical check.

C

Champion
A trained builder who helps colleagues in their team and passes feedback to the enablement lead. See champions.
Citizen developer, citizen development
The older IT-governance term for business staff building apps, usually on low-code platforms. Search term; this site says builder. See vibe coding vs. citizen development.
Citizen-developed (app)
An app built by business staff outside IT's delivery process. Many organizations, especially IT and governance teams, use it for what others call a vibe-coded app; the two are alternatives. Search term; this site says worker-built app. See vibe coding vs. citizen development.
Coding agent
An AI tool that edits code or runs commands with some autonomy; also called an agentic coding tool. See coding agents.
Compliance owner
The person or function accountable for audit-scope apps: change-management evidence, control testing and attestation scope. Not "auditor" for an internal role. See the audit-scope flag.
Context engineering
Curating everything an AI tool sees (instructions, files, retrieved content, tool descriptions), not just the prompt. Practitioners now treat it as the successor to "prompt engineering".1
Critical
The highest risk tier: "If it fails or errs, harm reaches customers, money, legal obligations or people's livelihoods." A business app owner plus IT or engineering as co-owner (graduation); IT-supported. See risk tiers.

D

Data classes
The four labels Q1 uses for data, from least to most sensitive: Public, Internal, Confidential and Restricted. The exact boundaries are in Q1; see also data handling.
Decision Record
A short, dated record of a decision: context, options, the choice, who decided and the review date. See the template.
Deny-by-default
Nobody can read or change data until a rule allows it. The opposite of public-by-default, the setting behind most documented exposures. See exposed data and secrets.
DORA metrics (the four keys)
Software delivery measures from the DORA research program, originally four: deployment frequency, lead time, change failure rate and time to restore. As of October 2026 DORA uses five, renaming the last "failed deployment recovery time" and adding deployment rework rate.2 See measurement.
Downgrade (owner-signed)
Placing an app one tier lower, by a signed Decision Record, when one answer is the only reason for the higher tier and compensating controls remove that reason; never for a floor or below a legal answer. Three standard downgrades are pre-defined. The rules and signers are on the downgrade path.

E

Enablement lead
The person who runs the central enablement function: intake and triage, the registry, training, champions and program metrics. Not "CoE lead". See roles and decision rights.
End-user computing (EUC)
Spreadsheets and tools business staff build outside IT's systems. EUC failures share today's root causes: no owner, no test, no independent review. See failure patterns.
Exception (bounded)
A change to when or how a gate is met, such as a runbook due after go-live, never to whether it applies. It needs a named approver, a written reason, an expiry of 90 days or less and an entry in the app record, and there are no exceptions to the floors or the legal triggers. See bounded exceptions.
Evals (eval set)
A fixed set of real inputs with expected outputs, including refusals and injection attempts, that an AI feature must pass, re-run on every model or prompt change. See evaluate an AI feature.
Exposure scan
An external check that an app's data, storage and URLs aren't reachable without sign-in. Not a penetration test unless it is one. See exposed data and secrets.

F

The five never-dos
(1) Never put restricted data into a tool that isn't approved for it. (2) Never put keys, passwords or tokens in an app's code, front end or prompts. (3) Never build or host a shared app on a personal account. (4) Never give an AI agent write access to live systems without a sandbox run first, a scoped credential and a restore you've tested. (5) Never turn on permission-bypass modes in coding agents. See the five never-dos.
Floor-breach rule
When an answer is higher only because the app breaks one of the two floors by mistake (say, a Team app's storage is publicly readable), contain the breach, record the as-found answer and a fix-by date within 5 working days, and tier the app as fixed. If the fix isn't confirmed by that date, the as-found answer applies. See the two floors.

G

Gate matrix
The table of what each risk tier requires at each lifecycle stage, from intake to retirement. It has one home: the gate matrix, also available as a CSV.
Graduation
Moving an app into shared ownership with IT or professional engineering: the business app owner stays, and IT or engineering becomes co-owner. It isn't a full handover by default, and not "handoff to IT". Every Critical app graduates; one that's Critical at intake plans graduation at intake. See graduate to IT.

I

Important
The third risk tier: "Outside users or a business process rely on it, or it holds Confidential data (most personal data), or it writes to a shared system of record." A business app owner with a named backup; assisted support. See risk tiers.
Intake and triage
The short form a builder completes before sharing an app or widening its data, reach or actions. It asks the six tier questions first. See intake.

J

Jurisdiction flag
An intake question, not a tier: does the app handle information about people in Quebec, the EU or the UK, or about staff who work in the EU or UK, or is an AI feature's output used in the EU? If you don't know, answer yes. Yes routes the app to counsel before building; it never changes the tier. Being the provider of a high-risk AI system under the EU AI Act is handled by this counsel route, not by a tier. See the jurisdiction flag.

K

Kill criteria
Conditions, set at intake or at a pilot's start, under which the app or pilot stops, such as no measured benefit by a date. See run a pilot.

L

Lethal trifecta
An agent or AI feature with private data, untrusted content and a way to send data out. With all three, prompt injection can steal the data.3 See the trifecta rule.
Low-code / no-code
Platforms for building apps by configuring forms, tables, flows and connectors, with little hand-written code. See low-code vs. AI app builders.

M

Maintainer
A person other than the original builder who changes the app. See change an app safely.
MCP (Model Context Protocol)
An open protocol connecting AI tools and agents to external tools and data. As of October 2026 the current specification is version 2026-07-28. It says hosts must get explicit user consent before invoking a tool, and admits the protocol itself can't enforce that, so the platform has to.4 See MCP and agents.
MCP server
A program that exposes tools or data to agents over MCP. Treat its tool descriptions as untrusted, and allow-list the servers you use. See MCP and agents.

N

Non-human identity
An account, key, token or grant an app or agent uses instead of a person's sign-in. Each needs a named owner, least privilege and revocation when the owner leaves.5

O

OAuth grant
Permission given to an app to act for a person or administrator in another system, such as a mailbox or CRM. Record and review grants. See MCP and agents.
Orphaned app
A worker-built app with no current owner. It often stays live with standing access to connected systems.6 See when the builder leaves.

P

Paved road
The supported, pre-approved way to build, host and run worker-built apps. Also called a golden path. See the paved road.
Prompt injection
Text the AI reads (a web page, document, ticket or tool description) carrying instructions the AI then follows. First in the OWASP Top 10 for LLM applications, 2025 edition.7 See AI feature security.

R

Retire
Shutting down an app and handling its data, access and grants. Not "decommission"; "sunset" only for whole programs. See retire an app.
Retire-by date
The date in the registry by which an app is renewed or retired. Sandbox apps default to 90 days. See retire an app.
Risk tier
The classification that sets the controls an app needs. Every worker-built app sits in one of four risk tiers (Sandbox, Team, Important or Critical), set by the highest answer to six questions about its data, users, internet reach, dependence, decisions about people and actions, and two floors apply at every tier. See risk tiers.
Row-level security
Database rules limiting which rows each user can read or change. Its absence, with a database key in the front end, is a common cause of exposed data. See exposed data and secrets.

S

Sandbox
The lowest risk tier: "A personal, time-boxed experiment. Only you use it, on public data or your own work files with no Confidential or Restricted data, and it's private by default." The builder; self-supported. See risk tiers.
SBOM (software bill of materials)
A machine-readable list of an app's components and dependencies. See the security checklist.
Secrets manager
A service that stores keys, passwords and tokens outside an app's code. See the security checklist.
Shadow AI
Workers using AI tools, or building apps with AI, that the business hasn't approved; this site says unsanctioned AI use. Search term.
The six tier questions
The intake questions that set an app's tier: Q1 data, Q2 users, Q3 reach, Q4 dependence, Q5 decisions about people, Q6 actions; the app takes the highest answer. The exact wording is on the six tier questions.
Slopsquatting
Publishing a malicious package under a name AI models tend to invent, so whoever installs the suggestion installs the attacker's code. See the security checklist.
SLSA
Supply-chain Levels for Software Artifacts: levels for build provenance and source integrity, at version 1.2 as of October 2026.8 See guardrails.
Spec-driven development
Writing a short specification first and having the AI build and test against it. Promising but early. See write a build spec.
Standing instructions (context files)
A file of rules an AI coding tool reads every session. Treat it as versioned, reviewed configuration. See a standing-instructions starter.
Support level
Who supports an app, set by its tier: self-supported, team-supported, assisted or IT-supported. Not "support tier". See the support model.
System of record
The shared business system that holds the authoritative copy of some data, such as a CRM, ERP, HR system or ticketing system. Writing to one is a Q6© answer. See the six tier questions.

T

Team
The second risk tier: "Shared inside the business (one team, several teams or company-wide), on internal data. If it stops, the people who use it work around it." A business app owner; team-supported (builder plus a champion). See risk tiers.
Technical check
At Important and above, an automated exposure scan and data-rule negative tests, plus a 30–45 minute check against a fixed list, done by a technical reviewer. See business review and technical check.
Technical reviewer
A platform or IT engineer, or a trained champion on the published list, who does the Important-tier technical check. See roles and decision rights.
Token passthrough
An MCP server accepting a token not issued to it and passing it on. The MCP specification forbids it.4
Tool permission
What an agent may do with a connected tool: read, write, delete, send or run commands. Grant the least that works. See MCP and agents.
The trifecta rule
The first of the two floors: "No agent or AI feature may combine private data, untrusted content and a way to send data out (the 'lethal trifecta'). Remove one leg before it runs, in Sandbox too."
The two floors
Two rules at every tier, with no exceptions or downgrade: (1) the trifecta rule; (2) no internet reach without company sign-in below Important. A mistaken breach is handled by the floor-breach rule. See the two floors.
Typical business
The site's audience: a business with an IT function and some security capability, without unlimited platform engineering, that handles some sensitive data. See getting started.

U

Unsanctioned AI use
Workers using AI tools the business hasn't approved; searchers say shadow AI. In 2026 vendor surveys, about half to two-thirds of office workers admitted it.9 See the unsanctioned AI use policy.

V

V.E.R.I.F.Y.
The builder's six-step self-check before release, and the site's one brand mnemonic: Validate the purpose; Exercise it with real examples; Restrict access; Inspect secrets and settings; Fact-check the output; Yield a recorded decision. See the V.E.R.I.F.Y. self-check.
Vibe coding
Building software by describing it to an AI and accepting what it generates, often without reading the code. Search term; this site says AI-assisted building. See what is vibe coding?
Vibe-coded (app)
An app built mostly by vibe coding. The press and security researchers tend to say "vibe-coded"; IT and governance teams more often say citizen-developed for the same thing. Search term; this site says worker-built app.

W

Worker-built app
An app, automation or agent built outside a funded engineering team's normal delivery process, by anyone, including professional developers building their own tools, with or without AI help. It includes automations, scheduled jobs, workflow-tool flows with AI steps, and agents built in SaaS agent builders or coding tools. See the lifecycle.

Don't: number the tiers

Never write the tiers as numbers or as low, medium and high. IT service management ranks a top-numbered system as the most critical, and help desks number their support levels, so numbers get misread. Say Sandbox, Team, Important and Critical, and support level for support.

Not legal advice

Not legal advice. This page summarizes laws as of October 2026 for a general audience. It was written with an AI legal-research persona and hasn't been reviewed by a lawyer for your situation. Ask your counsel.

These terms mean different things in different laws, so each entry names its jurisdiction. In North America, building an internal app with AI triggers few AI-specific duties; most come from what the app does. See US and Canadian laws and decisions about people.

Adverse action
US federal. Under Regulation B (Equal Credit Opportunity Act): a refusal to grant credit in substantially the amount or terms requested (unless the creditor counteroffers), a termination of an account or an unfavorable change in its terms that doesn't affect all or substantially all of a class of the creditor's accounts, or a refusal to increase credit on request (12 C.F.R. §1002.2(c)(1)).10 The creditor must give a statement of reasons that is "specific" and gives "the principal reason(s)" (§1002.9(b)(2)).11 The Fair Credit Reporting Act's own definition is broader: it also covers denying or worsening insurance and denying employment or any other employment decision that hurts a current or prospective employee (15 U.S.C. §1681a(k)).12 A business acting on a consumer report must give an adverse-action notice (§1681m), and an employer must first give a copy of the report and a summary of rights (§1681b(b)(3)).13 HR apps are the likeliest worker-built trigger.
AI system
EU AI Act (Art. 3(1)): "a machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments".19 For a worker-built app: one with an AI feature (an LLM call, for example) can be an AI system; code that an AI tool helped write is not, if the app itself doesn't infer its outputs. US and Canadian laws use their own terms, such as "automated decisionmaking technology" in California. See is an app built with AI an "AI system"?
Consequential decision
Colorado. Under SB 26-189 (signed 14 May 2026, effective 1 Jan 2027; attorney general rules pending), a decision affecting an individual's access to, eligibility for, or compensation related to education, employment, housing, financial or lending services, insurance, health care, or essential government services and benefits. Deployers of a covered automated decision-making technology (one that materially influences such a decision) must give notice at the point of interaction, a plain-language explanation within 30 days after an adverse outcome, and, on the person's request after an adverse outcome, meaningful human review and reconsideration. That review is a right the person invokes, not a standing duty to review every decision.14 California uses "significant decision" for a similar list, without insurance (Cal. Code Regs. tit. 11 §7001(ddd)).15 This site asks about these decisions in Q5 of the six tier questions.
Meaningful human review
This site's four-part test, adapted from the CCPA regulations' three-element definition of "human involvement" (Cal. Code Regs. tit. 11 §7001(e)(1)(A)–(C)): the reviewer (1) knows how to interpret the app's output, (2) reviews that output, (3) considers other relevant information, and (4) has the authority to change the decision. If any of the four is missing, the review doesn't count, and Q5 is answered ©.15 The California rules took effect on 1 Jan 2026, with automated-decisionmaking duties applying from 1 Jan 2027. The phrase "meaningful human review" itself comes from Colorado SB 26-189, where it's a review the person can request after an adverse decision.14 For California discipline and termination decisions, SB 947 adds a corroboration step. Not "human in the loop" as a vague claim. See what meaningful human review requires.
Personal information
California (Cal. Civ. Code §1798.140(v)): information that identifies, relates to, describes, is reasonably capable of being associated with, or could reasonably be linked, directly or indirectly, with a particular consumer or household. AB 1008 (2024) added that it can exist in "abstract digital formats", including AI systems capable of outputting personal information.2021 Canada, federal (PIPEDA s. 2): "information about an identifiable individual".23 Quebec (Law 25, s. 2): information that relates to a natural person and allows that person to be identified, directly or indirectly.16 On this site, Q1 asks about "information about identifiable people": business contact details and staff work-administration details are Internal, and most other personal information is Confidential. See data classes and data handling.
Privacy impact assessment
Quebec. Law 25 requires one "for any project to acquire, develop or overhaul an information system or electronic service delivery system involving the collection, use, communication, keeping or destruction of personal information" (s. 3.3), and another before communicating personal information outside Quebec (s. 17).16 So building a worker-built app that handles personal information is itself a trigger. EU: the GDPR's data protection impact assessment applies where processing is likely to result in a high risk (Art. 35).17
Provider and deployer
EU AI Act (Art. 3(3) and 3(4)). A provider "develops an AI system" or "has an AI system" developed (the definition also covers general-purpose AI models), "and places it on the market or puts the AI system into service under its own name or trademark". A deployer is "using an AI system under its authority", outside personal, non-professional use.19 Using a bought AI tool usually makes a business a deployer, but the Act reaches a deployer only if it has an EU establishment or location, or the tool's output is used in the EU (Art. 2(1)(b) and ©); the same reach applies to its AI-literacy measure (Art. 4). Putting a worker-built app with an AI feature into service can make the business a provider, if the app is put into service in the EU or its output is used there (Art. 2(1)); most provider duties apply only to high-risk systems or to the Art. 50 transparency cases. Code written with AI isn't itself an AI system. See the EU AI Act.
Sensitive personal information
California (Cal. Civ. Code §1798.140(ae)). Includes government ID numbers (Social Security, driver's license, state ID, passport); account log-ins and financial account or card numbers with the code or password that gives access; precise geolocation; racial or ethnic origin, citizenship or immigration status, religious or philosophical beliefs, union membership; contents of mail, email and texts not addressed to the business; genetic data; neural data (information generated by measuring the activity of a person's central or peripheral nervous system, not inferred from nonneural information; added by SB 1223, effective 1 Jan 2025); biometric data used to identify someone; and health, sex-life and sexual-orientation information.2022 The CPPA's regulations mirror the current list (§7001(bbb)), and processing it can require a risk assessment (§7150(b)).15 California gives consumers a right to limit its use (an opt-out); other US states use "sensitive data" for similar lists, usually with opt-in consent. On this site, most of these categories are Confidential (Q1©); other people's log-in credentials and full card numbers are Restricted (Q1(d)); and collecting biometric identifiers or consumers' health or other sensitive data outside HIPAA goes to counsel first.
Solely automated decision
A decision that takes effect with no meaningful human involvement; the test varies. EU: GDPR Art. 22 gives a right not to be subject to a decision "based solely on automated processing, including profiling" with legal or similarly significant effects, subject to exceptions with safeguards.17 UK: since 5 Feb 2026, Arts. 22A–22D permit such decisions with safeguards (information, representations, human intervention, contest), but decisions based on special-category data stay restricted (Art. 22B).18 Quebec: a decision "based exclusively on automated processing" requires notice and a chance to submit observations to a staff member who can review it (s. 12.1).16 California: the CCPA rules (effective 1 Jan 2026; automated-decisionmaking duties from 1 Jan 2027) cover technology that replaces or substantially replaces human decisionmaking.15 On this site, an outcome that can take effect before meaningful human review is Critical under Q5.

Former framework names

Earlier versions of this site used eight acronym frameworks. Seven are folded into plainly titled pages, and their old URLs redirect. V.E.R.I.F.Y. is kept, with a new expansion.

Former name Now lives at What survives
S.C.A.F.F. Write a build spec A five-part spec, with a non-developer example; the prompts move to starter prompts
V.E.R.I.F.Y. The V.E.R.I.F.Y. self-check Kept as the one brand mnemonic, with the expansion in the V.E.R.I.F.Y. entry
C.L.E.A.R. Business review and technical check The independent-review checklist, as a business review plus a technical check, with code and no-code variants
S.E.C.U.R.E. Security checklist The per-change security checks, merged into one checklist
S.H.I.E.L.D. Guardrails The enforcement-placement matrix
R.E.F.A.C.T. Change an app safely "Add a test first, change in small steps"
D.O.C.S. Document your app Doc ownership; what AI may and may not draft
C.O.D.E.S. Enablement and training Charter, onboarding and sync, as enablement and community

Don't: keep using the old expansions

The old acronyms had conflicting expansions; V.E.R.I.F.Y. alone had three. Use the page names above and the one V.E.R.I.F.Y. expansion in this glossary.

Questions to ask your counsel

The counsel questions behind these legal terms are on US and Canadian laws for AI-built apps. Start with which privacy laws reach the people your apps touch, whether any app makes or influences a consequential or significant decision, and what counts as meaningful human review for you.

Assumptions and evidence

This page relies on no adoption assumptions of its own; each canonical page lists the assumptions behind its term. Evidence: legal and standards definitions link to the statute, regulation or standard itself, and the few practitioner or vendor sources are labelled. Facts on this page are as of October 2026.

  1. Thoughtworks Technology Radar, "Context engineering" (Adopt), Apr 2026 [primary], https://www.thoughtworks.com/radar/techniques/context-engineering, accessed 2026-10-21. ↩

  2. DORA, guide to its software delivery performance metrics [research], https://dora.dev/guides/dora-metrics-four-keys/, accessed 2026-10-21. ↩

  3. Simon Willison, "The lethal trifecta for AI agents", 16 Jun 2025 [primary], https://simonwillison.net/2025/Jun/16/the-lethal-trifecta/, accessed 2026-10-21. ↩

  4. Model Context Protocol, specification version 2026-07-28 and its security best practices [primary], https://modelcontextprotocol.io/specification/2026-07-28, accessed 2026-10-21. ↩↩

  5. OWASP, "Non-Human Identities Top 10", 2025 [primary], https://owasp.org/www-project-non-human-identities-top-10/, accessed 2026-10-21. ↩

  6. OWASP Citizen Development Top 10, "CD-SEC-09: Asset Management Failures" [primary], https://owasp.github.io/www-project-citizen-development-top10-security-risks/content/2022/en/CD-SEC-09-Asset-Management-Failures, accessed 2026-10-21. Guidance from 2022; the orphan risk it describes persists. ↩

  7. OWASP, "Top 10 for LLM Applications", 2025 edition [primary], https://genai.owasp.org/llm-top-10/, accessed 2026-10-21. ↩

  8. SLSA, "What's new in SLSA v1.2" [primary], https://slsa.dev/spec/v1.2/whats-new, accessed 2026-10-21. ↩

  9. PagerDuty, survey release on unauthorized AI tool use, 11 Jun 2026 [vendor], https://s206.q4cdn.com/635206389/files/doc_news/PagerDuty-Report-Finds-Two-Thirds-66-of-Office-Professionals-Have-Used-Unauthorized-AI-Tools-at-Work-2026.pdf; and The Register on an Okta survey, 27 May 2026 [news], https://www.theregister.com/ai-ml/2026/05/27/bosses-blinded-by-confidence-about-shadow-ai-use-by-workers/5247275, accessed 2026-10-21. ↩

  10. Consumer Financial Protection Bureau, Regulation B, 12 C.F.R. §1002.2 [statute], https://www.consumerfinance.gov/rules-policy/regulations/1002/2/, accessed 2026-10-21. ↩

  11. Consumer Financial Protection Bureau, Regulation B, 12 C.F.R. §1002.9 [statute], https://www.consumerfinance.gov/rules-policy/regulations/1002/9/, accessed 2026-10-21. ↩

  12. 15 U.S.C. §1681a, Fair Credit Reporting Act definitions [statute], https://uscode.house.gov/view.xhtml?req=granuleid:USC-prelim-title15-section1681a&num=0&edition=prelim, accessed 2026-10-21. ↩

  13. 15 U.S.C. §1681m, Fair Credit Reporting Act [statute], https://uscode.house.gov/view.xhtml?req=granuleid:USC-prelim-title15-section1681m&num=0&edition=prelim, accessed 2026-10-21. ↩

  14. Colorado General Assembly, SB 26-189 [statute], https://leg.colorado.gov/bills/sb26-189, accessed 2026-10-21. ↩↩

  15. California Privacy Protection Agency, approved regulations on automated decisionmaking technology, risk assessments and cybersecurity audits, effective 1 Jan 2026 [regulator], https://cppa.ca.gov/regulations/pdf/ccpa_updates_cyber_risk_admt_appr_text.pdf, accessed 2026-10-21. ↩↩↩↩

  16. Act respecting the protection of personal information in the private sector, CQLR c. P-39.1 (Quebec Law 25) [statute], https://www.legisquebec.gouv.qc.ca/en/document/cs/P-39.1, accessed 2026-10-21. ↩↩↩

  17. Regulation (EU) 2016/679 (GDPR) [statute], https://eur-lex.europa.eu/eli/reg/2016/679/oj, accessed 2026-10-21. ↩↩

  18. Data (Use and Access) Act 2025, s. 80 [statute], https://www.legislation.gov.uk/ukpga/2025/18/section/80, and SI 2026/82 [statute], https://www.legislation.gov.uk/uksi/2026/82/made, accessed 2026-10-21. ↩

  19. European Commission AI Act Service Desk, "Article 3: Definitions" [primary], https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-3, accessed 2026-10-21. ↩↩

  20. California Civil Code §1798.140, current text [statute], https://leginfo.legislature.ca.gov/faces/codes_displaySection.xhtml?lawCode=CIV&sectionNum=1798.140, accessed 2026-10-21 (our fetcher was blocked; the CPPA regulations above mirror the current definitions). ↩↩

  21. California Legislature, AB 1008 (2024) [statute], https://leginfo.legislature.ca.gov/faces/billNavClient.xhtml?bill_id=202320240AB1008, accessed 2026-10-21. ↩

  22. California Legislature, SB 1223 (2024), neural data [statute], https://leginfo.legislature.ca.gov/faces/billNavClient.xhtml?bill_id=202320240SB1223, accessed 2026-10-21. ↩

  23. Personal Information Protection and Electronic Documents Act, S.C. 2000, c. 5, s. 2 [statute], https://laws-lois.justice.gc.ca/eng/acts/P-8.6/page-1.html, accessed 2026-10-21. ↩