May this actor use, move, expose, or delete this data at this scope?
A valid identity or token may reach data. That does not authorize every use of it. Authority Control evaluates defined data actions such as export, deletion, publication, reclassification, or training use before consequence occurs.
A valid token can reach a production database.
That does not mean the actor has authority to delete it.
- 01Valid token reaches the database.
- 02System access is treated as authority.
- 03The deletion executes.
- 04Recovery begins after the data is already gone.
- 01Valid token reaches the database.
- 02Authority is checked at the point of deletion.
- 03The out-of-scope command is blocked or deferred.
- 04A decision record is created.
The same question applies to trades, payments, deployments, contracts, model changes, and data releases. AI did not create this failure mode. AI agents multiply it.
Access controls have grown more sophisticated every year. The scale of consequence has grown faster.
Access controls have grown more sophisticated every year. The scale of consequence has grown faster.
What the identity can reach
What the identity is allowed to produce
Reaching the data and creating consequence through it are different questions.
Bulk extraction beyond need
Changing sensitivity or access tier
Irrecoverable removal of records
Widening retrieval or query boundaries
Release into an operational or external system
Authorization for model training or fine-tuning
Where Authority Control operates: the organization that defines data authority, owns the data, and controls the integrations that move it holds the defining role. Authority Control enforces what the customer has defined.
This architecture applies at the customer edge against existing data exposures, without waiting for platform vendors. Every SaaS integration, every authorized token, every automated pipeline that moves data is a commitment surface where the authority check can be applied.
Access controls pass. Authority over consequence remains open.
Access is verified. Commitment is not. Four data cases follow the pattern.
+OAuth Integration Data Exfiltration Salesloft / Drift, 2025
OAuth tokens used for a legitimate Drift-to-Salesforce chatbot integration were used to extract Salesforce data across 700+ customer organizations. Every enterprise that consumes OAuth-mediated SaaS integrations carries this structural exposure today.
Zero Trust verifies the token and session, and confirms that the integration is permitted to reach Salesforce.
- The token is authenticated.
- The session is permitted.
- The integration can reach Salesforce.
Deployment: The Salesforce customer deploys and configures the authority policy. No platform-vendor cooperation is required for this scenario.
+Catastrophic-Scale Exfiltration Change Healthcare, 2024
Attackers accessed Change Healthcare systems and exfiltrated data for 192.7 million individuals, including medical data, SSNs, and insurance information. Every healthcare processor, payer, and clearinghouse carries this exposure today: authenticated access to processing systems can still create catastrophic data-movement consequence unless bulk extraction is evaluated as a distinct authority class before execution.
Zero Trust verifies credentials, network paths, and application access to the processing environment.
- The identity is authenticated.
- The application path is permitted.
- The system can technically export the data.
+API Misconfiguration TransUnion, 2025
Attackers exploited misconfigured API permissions in a third-party Salesforce integration, exfiltrating SSNs and credit data for 4.4 million customers. Every enterprise running third-party APIs and SaaS integrations carries this exposure today. A path that serves one legitimate lookup can also serve millions of records unless lookup and bulk extraction are governed as different commitment classes.
Zero Trust verifies the third-party integration, the Salesforce environment, and the API path.
- The integration is recognized.
- The API path is available.
- A legitimate lookup and a bulk query may use the same technical route.
+Insider Data Exfiltration Coinbase, 2025
Overseas support contractors used legitimate access to exfiltrate customer data for 69,461 users, including partial SSNs and ID images. Their data interactions were structurally indistinguishable from legitimate support operations. Estimated cost: $400 million. Least privilege was satisfied. Least authority was absent.
Zero Trust verifies the contractor identity, role-based access, and application path used for support work.
- The contractor is authenticated.
- The support application is permitted.
- Viewing, copying, and exporting may look similar at the access layer.
Access narrows reach. Authority narrows consequence. Together, they limit blast radius across two independent dimensions.
- 1. ITRC, 2025 Annual Data Breach Report (20th edition), released January 29, 2026; 3,322 compromises and 278,827,933 victim notices. Prior editions: 2023 (3,205 compromises), 2024 (3,152 compromises).
- 2. IBM Security and Ponemon Institute, Cost of a Data Breach Report 2025; average U.S. breach cost of $10.22M.
- 3. Compiled from SEC filings, FTC enforcement actions, and company breach disclosures.
- 4. Anthem Inc., breach notification filing, February 2015; HHS Office for Civil Rights.
- 5. U.S. Office of Personnel Management, cybersecurity incident disclosures, 2015.
- 6. Equifax breach disclosures and FTC settlement, 2017–2019.
- 7. UnitedHealth Group and Change Healthcare breach disclosures, 2024.
- 8. National Public Data breach disclosures and reporting, 2024.
Governing data consequence starts with one integration.
Start with the surfaces where valid access already moves data: SaaS integrations, tokens, and automated pipelines.