What Happens After Every Company Becomes a Data Company?
Art Anthony · rss · 2026-07-28 source ↗ Reader 🖥️
Agent Governance (Internal)Agentic AIAI Regulation & Governance (External)API Products & SprawlConsumer Data Stewardship & MinimizationPrivilege Drift / Scope Sprawl
As every company becomes a data company and AI agents begin consuming that data autonomously, organizations must abandon convenience-first identity patterns — broad SSO, single super-profiles, session-long OAuth grants — in favor of a Zero-Trust posture of least-privilege, just-in-time, minimally-scoped, continuously-evaluated access for humans and agents alike, treating themselves as active data stewards rather than passive collectors.
practical · argumentative — A mixed expository/argumentative industry blog post: it opens with descriptive trend-setting (Fanatics, Coca-Cola, the 2018 Forbes thesis) and closes with a normative architectural prescription (Zero-Trust identity and access design for both human users and AI agents) aimed at API/IAM architects and security engineers.
Structure
| Part | Role |
|---|---|
| Opening: the 2018 Forbes 'every company is a data company' thesis, updated with the Fanatics case (16 data attributes/customer, billions of data points/year) | Establishes the empirical premise — data-centrality is now near-universal, not a tech-sector niche |
| Responsible Stewardship of Customer Data: 'how much data is too much', SSO reconsidered, the zero-trust question checklist for agents/users | Problem framing — names the risk (attractive target, regulatory exposure) and introduces standing-privilege as the thing to eliminate |
| Going Beyond OAuth: domain separation, token minimization, fine-grained access, continuous risk evaluation, decentralized identity; OAuth 2.0's 2012-era design vs. agentic access; JIT authorization and Rich Authorization Requests | Core prescription — the concrete architectural alternative to broad-grant identity |
| User-side counter-pressure: VPNs, Apple Hide My Email as evidence users are already self-enacting minimization | Corroborating evidence — shows the prescribed shift isn't only top-down but matches user behavior/demand |
| The Ethics of Data Minimization: Target pregnancy-prediction story, Coca-Cola Freestyle vending-machine data | Narrative illustration — reframes the question from 'can we build this profile' to 'should we', and establishes that stakes scale with domain sensitivity (soda preference vs. health/finance/gambling) |
| AI, Compartmentalization, and the Future of Identity: EU AI Act, gambling-industry AI ethics committees, California CPRA/ADMT rulemaking | Regulatory-convergence evidence — shows external law is about to force the architectural change the article prescribes, while also casting AI as a governance tool (fraud detection, verification) and not only a governance risk |
| Closing restatement: API infrastructure and Zero-Trust access as the operating model for the data-company era | Conclusion — ties the architectural prescription back to the opening thesis |
Key terms
- data steward Consumer Data Stewardship & Minimization
An organization's active, ongoing ethical duty to protect and limit the customer data it holds, scaled to that data's sensitivity — a self-conception the article urges companies to adopt, not merely a compliance posture. - zero trust (applied to identity)
Not network segmentation but a checklist of continuous questions asked of every access request by a human or AI agent: how much data is needed, for how long, what is the intent, and should any standing privilege exist at all. - standing privilege
Access that persists indefinitely once granted — via SSO or a unified profile — rather than being issued only for the duration and scope of a specific need; something the article implies should generally not exist for agents or, increasingly, for humans. - agentic access
The pattern of an AI agent, rather than a human, making authenticated API calls on a user's or organization's behalf — treated as the forcing function that makes 2012-era OAuth 2.0 inadequate and demands its evolution. - super-profile Consumer Data Stewardship & Minimization
A single, unified, cross-service customer profile an organization holds for convenience, marketing, and operational streamlining — framed by the author as increasingly incompatible with sound identity architecture. - data minimization Consumer Data Stewardship & Minimization
Restricting what data or identity attributes are collected, retained, or exposed to only what a specific system or request genuinely needs — operationalized via short-lived tokens, selective disclosure, and pseudonymization, not just a legal checkbox.
Propositions
- explicitEvery company is now effectively a data company; a decade after the 2018 Forbes thesis, virtually all organizations, in and beyond tech, prioritize data-driven decisions.conf: high
- explicitcorroboratedThe more customer data a company holds, the greater its responsibility to protect it, since rich data holdings make a company both a more attractive target for bad actors and subject to more intensive regulatory scrutiny.conf: highsupports ←Responsible AI Is Becoming a Growth Strategy · p8 — This article's consumer-trust deficit statistic (93% AI deployment vs. 23% consumer trust) supports the other document's claim that the more customer data/AI-driven decisioning a company holds, the greater its responsibility to earn and protect trust around it.
- explicitSingle sign-on no longer fits newer definitions of identity in ecosystems of connected services, because one login can span both low-stakes actions (buying a jersey) and high-stakes ones (depositing money to gamble) under a single parent company.conf: high
- explicitZero-trust questions — how much data/access is needed, for how long, what is the intent, and whether any standing privilege should exist — are especially relevant to AI agents now interfacing with sensitive data.conf: high
- explicitOrganizations should adopt data-domain separation, token minimization, fine-grained access controls, continuous risk evaluation, and decentralized identity with selective disclosure in place of broad, single-authentication access.conf: highsupportsHow to Manage Privilege Drift in Multi-Agent Systems · p6 — Converges independently on the same JIT/short-lived-credential prescription: privilege-drift approaches it from the angle of agent permission accumulation over time, this article from the angle of customer-identity and OAuth architecture.
- explicitcorroboratedOAuth 2.0, designed in 2012 before autonomous non-human API consumption was a consideration, must be adapted — via just-in-time authorization and Rich Authorization Requests — to enforce least privilege for agentic access.conf: highextendsWhy Token Handling Must Evolve for AI Agents · E5 — Both argue for short-lived, minimally-scoped, just-in-time tokens over broad session grants; this article generalizes the prior piece's agent-token-specific prescription into a fuller identity-architecture agenda (domain separation, decentralized identity, selective disclosure) spanning both human SSO and AI-agent OAuth access, and explicitly self-references 'a recent post' covering JIT authorization.supports ←What Is a Token? A Developer's Guide to Every Type · p5 — This article's account of why OAuth 2.0 and OpenID Connect became the standard, scalable basis of token-based auth is the premise the other piece complicates: those same standards, designed in 2012 for human sign-in, are argued to now need adaptation (just-in-time authorization, Rich Authorization Requests) for autonomous non-human API consumers.
- explicitPlacing the burden of opting out of tracking on end users, rather than organizations handling data responsibility internally, is a mistake — evidenced by users independently adopting VPNs and tools like Apple's Hide My Email to anonymize their own data.conf: medium
- explicitcorroboratedThe ethical question for customer profiling is not whether organizations can build cross-product, cross-device profiles of users, but whether they should, since not every system needs every attribute and not every API request needs a full identity profile.conf: highsupports ←This AI notetaker won't sell surveillance to your boss · p1 — Granola's refusal to give employers full-transcript access is a concrete real-world instance of the abstract 'can vs. should' data-ethics distinction — not every stakeholder that can access data should.
- implicitAI's introduction into data practices is accelerating the shift of data ethics from voluntary internal discussion to binding regulation, converging AI-specific law (EU AI Act, California ADMT rulemaking under CPRA) with existing data-protection regimes (GDPR, CCPA).conf: medium
- implicitAI agents are simultaneously a source of new data-governance risk (hyper-targeted, real-time exploitation of profiles) and a legitimate governance tool in their own right (fraud prevention, automated document verification, detecting problem-spending patterns) — the same agentic capability cuts both ways.conf: medium
Arguments
Organizations should replace single super-profiles, SSO, and broad OAuth grants with domain-separated, minimized, continuously-evaluated, fine-grained access for both humans and AI agents. strength: medium
Explicit premises:
- Consolidated customer data at scale (Fanatics: 16 attributes/customer, billions of data points/year) increases attractiveness to bad actors and regulatory exposure
- Ecosystems increasingly mix low-stakes and high-stakes actions under one identity/login
- Zero-trust questions (need, duration, intent, standing privilege) are especially urgent for AI agents accessing sensitive data
- Regulatory convergence (EU AI Act, CPRA/ADMT, GDPR/CCPA) is tightening data-handling requirements
Implicit premises:
- The liability/compliance reduction from minimization outweighs the marketing and convenience value organizations currently extract from unified profiles
- Fine-grained architectural separation is technically and commercially feasible without crippling personalization or streamlined operations
OAuth 2.0 must evolve — via just-in-time authorization and Rich Authorization Requests — to support least privilege for AI-agent (non-human) consumption of APIs. strength: medium
Explicit premises:
- OAuth 2.0 debuted in 2012, before autonomous API consumption by non-humans was a design consideration
- SSO made human authentication easier but did not ensure it was handled responsibly
- A prior post (same publication) already described JIT authorization and Rich Authorization Requests as ways to model non-human access
Implicit premises:
- The maturity of JIT/RAR tooling is sufficient to deploy in production without unacceptable latency or operational overhead — left unexamined
Narratives
- Fanatics' transition from a brick-and-mortar Florida sporting-goods store to a company holding up to 16 data attributes per customer and billions of data points annually, per CEO Michael Rubin and Principal Product Manager Daniel Fox.
→ Grounds the abstract 'every company is a data company' thesis in a concrete, non-tech-native business, making the scale and normalcy of the shift tangible before the article turns to its risks. - A New York Times Magazine account of Target marketers inferring a teenage girl's pregnancy from her shopping habits and mailing baby-product coupons that landed on her unaware father's desk.
→ Illustrates that data-driven inference can be statistically correct yet ethically and socially harmful, reframing the article's central question from 'can organizations build this profile' to 'should they' — the pivot for the whole 'Ethics of Data Minimization' section. - Coca-Cola's admission that Freestyle vending-machine data (roughly 11 million servings/day, per VP Megan Tallman) informs future drink development.
→ Serves as a low-stakes contrast case — deliberately 'fairly innocuous' — to set up the argument that the same data-driven logic becomes far higher-stakes in regulated domains like finance, healthcare, or gambling.
Questions addressed
- ✅ Why does holding more customer data create more organizational responsibility?
Richer data holdings make a company both a more attractive target for bad actors and subject to heavier regulatory scrutiny. - ✅ Is single sign-on still fit for purpose in an AI-agent, multi-ecosystem era?
No — SSO no longer fits newer definitions of identity when one login spans both benign and high-risk actions across a connected ecosystem. - ✅ How should OAuth adapt for AI agents?
Via just-in-time authorization, Rich Authorization Requests, and a general shift toward least-privilege, fine-grained access. - ⭕ How much customer data is too much?
The author poses this explicitly ('which begs the question of how much data is too much data') but never returns to answer it with a framework or threshold. - ⭕ Should AI agents (or humans) ever hold standing privilege?
The zero-trust question list raises this directly, but the article never states a general answer or decision rule — it leaves the question open as something organizations must interrogate case by case.
Gaps
- incompleteThe article never analyzes the tradeoff it implicitly creates: domain separation, token minimization, and decentralized identity reduce the very convenience, personalization, and cross-sell value that make unified super-profiles commercially attractive in the first place. No cost/benefit accounting or case study of an organization that made this shift is offered.
Read next: Read alongside CRM/marketing-attribution literature on the revenue value of unified customer profiles, to weigh against the security/compliance benefits this article argues for. - incompleteJust-in-time authorization and Rich Authorization Requests are proposed as the mechanism for agentic access, but the article never asks who authenticates the AI agent itself, or what secures the token-issuing/authorization service — the same structural gap already flagged elsewhere in this corpus for a companion Nordic APIs piece on agent token handling and for privilege-drift mitigation in multi-agent systems.
Read next: extract-token-handling-ai-agents#g2 and extract-privilege-drift-multi-agent-systems#g2 raise the identical open problem — agent identity is treated as solved rather than examined. - incompleteThe article's own central provocative question — 'how much data is too much data' — is posed rhetorically and never answered; no threshold, framework, or decision procedure is offered for organizations trying to operationalize 'minimization' in practice.
Read next: Data-minimization-by-design frameworks (e.g. NIST privacy engineering, GDPR Art. 5 guidance) that attempt to convert the principle into an auditable practice. - incompleteThe piece names EU AI Act, California CPRA/ADMT rulemaking, and GDPR/CCPA as converging pressures but never addresses what a multinational organization does when these regimes conflict (e.g. divergent definitions of 'automated decision', differing consent/legitimate-interest bases, or data-residency requirements) rather than simply layering.
Read next: Comparative privacy-regulation literature on cross-jurisdictional conflict (GDPR vs. CCPA/CPRA vs. sectoral US rules) for organizations operating across the EU and US simultaneously. - illogicalThe article treats decentralized identity with selective disclosure as a straightforward replacement for SSO without acknowledging the friction cost to users — but the article itself earlier credits SSO's whole value proposition as 'reducing friction and simplifying identity management.' The prescription is in unacknowledged tension with the article's own account of what made SSO attractive.
Read next: UX research on consent-fatigue and friction in selective-disclosure/decentralized-identity flows, to test whether the security gain the article prescribes is deliverable without reintroducing the friction SSO was built to remove.
Related documents
- 5 Signs That Authenticated API Traffic Is Actually Malicious
- Responsible AI Is Becoming a Growth Strategy
- How AI Swarms Are Disrupting Democracy
- Presentation: Autonomous Data Products for the Autonomous Era: Rethinking Data Architecture for GenAI
- AWS Launches Amazon GuardDuty Investigation Agent to Automate Threat Triage
- Building AI Agents For The Enterprise
- Why Developer Portals Must Evolve in the AI Agent Age
- Entur AS' innlegg — KI-løypa: Slik vurderer og innfører vi KI-verktøy i Entur
- Indirect Prompt Injection Exploits GitHub's AI Agent to Leak Private Repository Data
- Beyond Zero: Google Publishes Successor to BeyondCorp