PADFA Guide: FTC Data Broker Ban, Covered Data and Compliance Steps

A US company that sells or licenses data it did not collect itself can break federal law with a single delivery of sensitive data to a buyer controlled from China, Iran, North Korea or Russia, and there is no volume threshold to hide behind. The Protecting Americans' Data from Foreign Adversaries Act of 2024 (PADFA, sometimes written PADFAA), codified at 15 U.S.C. 9901, is a short statute enforced by the Federal Trade Commission (FTC). It is narrower than the Department of Justice (DOJ) Data Security Program in who it covers, but broader in what data counts and stricter in having no bulk minimum. This guide explains who counts as a data broker, the 17 categories of sensitive data, the 20% control test, how the FTC enforces the Act, how it differs from 28 CFR Part 202, and how to turn it into buyer screening and access controls, including for API keys and AI agents. It is written for privacy, security and data platform leads at any business that monetises third-party data.

Status line (checked 14 November 2026): PADFA took effect on 24 June 2024, 60 days after enactment on 24 April 2024. On 9 February 2026 the FTC announced warning letters to 13 data brokers reminding them of their PADFA obligations; the FTC press release reports no enforcement action, and none was found in the FTC sources checked for this guide. There is no implementing FTC rule: the statute itself sets the obligations.

Key takeaways

• It is a flat ban, not a compliance regime. Data brokers may not sell, license, transfer, give access to or otherwise make available personally identifiable sensitive data of a US individual to a foreign adversary country or an entity it controls. There are no licences, security conditions or bulk thresholds.
• Data broker is defined by what you sell, not your label. Any entity that, for valuable consideration, transfers data of US individuals it did not collect directly from them to an entity that is not acting as its service provider is a data broker, subject to five exclusions.
• Sensitive data is wide. Seventeen categories, including government identifiers, precise geolocation, log-in credentials, online activity tracked across sites over time, information about anyone under 17 and military service status.
• The control test bites at 20%. A buyer is controlled by a foreign adversary if it is based in or organised under the laws of China, Iran, North Korea or Russia, is at least 20% owned by such persons, or is directed or controlled by them.
• The FTC is signalling enforcement. Violations are treated as violations of an FTC trade regulation rule, so civil penalties are available from a first violation; the February 2026 warning letters cite up to USD 53,088 per violation.

Scope: who and what is covered

PADFA regulates one actor (the data broker), one act (making data available) and one class of recipient (foreign adversary countries and entities they control). The definitions in 15 U.S.C. 9901(c) do most of the work. This guide does not repeat the DOJ rule; for 28 CFR Part 202 scope, bulk thresholds and the CISA security requirements, see the DOJ Data Security Programme Guide.

Data broker and its five exclusions

A data broker is an entity that, for valuable consideration, sells, licenses, rents, trades, transfers, releases, discloses, provides access to or otherwise makes available data of US individuals that it did not collect directly from them to another entity that is not acting as a service provider. An entity is not a data broker to the extent it:

1. transmits an individual's data, including communications, at that individual's request;
2. provides a product or service where the data, or access to it, is not the product or service itself;
3. reports or publishes news or information concerning local, national or international events or other matters of public interest;
4. reports, publishes or otherwise makes available information already available to the public from public sources such as books, directories, broadcasts, news media or websites available without restriction; or
5. is acting as a service provider.

Service provider

A service provider collects, processes or transfers data on behalf of, and at the direction of, an individual or entity that is not itself a foreign adversary country or controlled by one, or a government entity, and receives that data from or on behalf of that individual or entity. So a processor working for a US client is not brokering, but the shelter fails if its principal is adversary-controlled.

Personally identifiable sensitive data

The Act covers sensitive data that identifies, or is linked or reasonably linkable, alone or in combination with other data, to an individual or to a device that identifies or is linked or reasonably linkable to an individual. Pseudonymous device-level records are therefore in scope if they can be linked back. The 17 categories of sensitive data are:

Category (Q) is an anti-evasion clause: selling innocuous-looking variables so the buyer can infer a covered category is still a violation.

Foreign adversaries and control

Foreign adversary country means a country specified in 10 U.S.C. 4872. The statute cross-refers to section 4872(d)(2), but that list now sits at 10 U.S.C. 4872(f)(2) following later amendments, and it names China, Iran, North Korea and Russia, as the FTC's February 2026 letters also state. An entity is controlled by a foreign adversary if it is:

• Based there: a foreign person domiciled in, headquartered in, with its principal place of business in, or organised under the laws of a foreign adversary country;
• Owned at 20%: an entity in which such a person or persons directly or indirectly own at least a 20% stake; or
• Directed: a person subject to the direction or control of either of the above.

The protected population is a United States individual: a natural person residing in the United States. Nationality is irrelevant; residence decides.

Key obligations

The prohibition in 15 U.S.C. 9901(a) is the only operative duty. It is unlawful for a data broker to sell, license, rent, trade, transfer, release, disclose, provide access to or otherwise make available personally identifiable sensitive data of a US individual to any foreign adversary country or any entity controlled by a foreign adversary. In practice that one sentence produces five working obligations:

1. Know which products are brokering: classify each data product and feed against the definition and exclusions, and record the reasoning.
2. Know which fields are sensitive: tag every field and derived segment against categories (A) to (Q), including inferred and modelled attributes.
3. Know the buyer: establish, before access and on change of ownership, whether each customer, reseller and their ultimate parent is controlled by a foreign adversary under the 20% and direction tests.
4. Control every access path: "provide access to" reaches API keys, data clean room seats, bulk file drops, cloud storage shares and self-service portals, not just negotiated sales.
5. Stop downstream leakage: the Act has no express onward-transfer clause, so contract and technical controls on resellers are how a broker shows it did not make data available to an adversary indirectly.

How PADFA differs from the DOJ Data Security Program

Both regimes can apply to one sale. A sale of a few hundred records of military status to a 25%-Russian-owned firm is likely below every Part 202 bulk threshold and outside its 50% covered person test, yet squarely within PADFA. Plan for the stricter element of each.

Timeline and current status

The FTC did not allege that the letter recipients broke the law, and said it is monitoring the marketplace. Watch for a first complaint, any change to the list in 10 U.S.C. 4872, and inflation adjustments to FTC civil penalty amounts.

Enforcement and penalties

• Legal basis: a violation is treated as a violation of a rule defining an unfair or deceptive act or practice under section 18(a)(1)(B) of the FTC Act (15 U.S.C. 57a(a)(1)(B)). The FTC enforces PADFA with the same means, jurisdiction, powers and duties as under the FTC Act.
• Civil penalties from the first violation: because PADFA violations are treated as rule violations, the FTC can seek civil penalties under section 5(m)(1)(A) without first obtaining a cease and desist order. The February 2026 warning letter template gives the figure as up to USD 53,088 per violation. The amount is adjusted for inflation each year, and because the cap is per violation, exposure grows with the number of violations alleged.
• Injunctions: the letters also note that the FTC can seek court injunctions.
• No private right of action: the Act itself gives individuals no claim, although law firm commentary notes plaintiffs and state attorneys general are pursuing related foreign data transfer theories under other laws.
• Other FTC powers preserved: nothing in PADFA limits the FTC's authority under any other law, so conduct can be charged as both a PADFA violation and an unfair practice under section 5.

What it means for AI agents and non-human identities

"Provide access to" is the phrase that turns PADFA into an identity problem. Most modern data products are delivered through credentials, not shipments.

• API keys and OAuth clients are the sale: each data API key or OAuth client issued to a customer is the access path the Act regulates. Tie every key to a screened legal entity, never to a sign-up email, and revoke keys automatically when a customer's control status changes or screening lapses. See the Service Account Security Guide.
• Shared storage and clean rooms: cross-account cloud storage grants, data share objects and clean room seats are standing access. Inventory them as customer entitlements with an owner and an expiry. See the Third-Party Access Guide.
• Resold credentials: category (H) makes account log-in credentials and access codes sensitive data in their own right. Any business that aggregates leaked credentials, session tokens or API secrets for resale must treat that inventory as covered. See the Secrets Management Guide.
• AI agents as buyers: an agent querying a data API acts under its operator's credentials. If that operator is controlled by a foreign adversary, the access is a PADFA question regardless of whether a human ever sees the output. Bind agent and machine credentials to the screened entity and log the purpose of each query. See the AI Agent Identity vs Service Account.
• AI training data deals: licensing a data set that includes covered categories to a model developer is making data available. Screen model and AI service providers for control like any other buyer, including their parent companies.

Compliance steps

1. Scope the business: list every product, feed, segment and data share. For each, record whether it transfers data the company did not collect directly, for consideration, to a non-service provider, and which exclusion applies, if any. Revisit when a product changes.
2. Tag sensitive data: map each field and derived segment to categories (A) to (Q). Treat modelled attributes that reveal a category (for example an inferred religion score) as that category. See the Data Classification Guide and the Data Mapping and ROPA Guide.
3. Build a control screen: for each customer and reseller, collect ultimate beneficial ownership, place of organisation, headquarters and principal place of business, and test against the four countries and the 20% threshold. Re-screen on renewal, on ownership change notices and at least annually.
4. Gate access technically: issue data credentials only after screening passes, attach the screening record to the credential, and block delivery to entities that fail. Pilot the gate on new customers first, then backfill existing ones in risk order (sensitive categories and higher-risk geographies first).
5. Contract for downstream control: add warranties that the buyer is not controlled by a foreign adversary, a ban on onward transfer to such entities, notice of ownership change and audit rights. The contract does not move the liability.
6. Handle failures (rollback): when a customer fails screening, suspend credentials first, preserve logs of what was delivered and when, then decide with counsel on termination and whether self-reporting is appropriate.
7. Align with Part 202: run the PADFA screen and the DOJ screen from one counterparty register, using the stricter test for each element: 20% ownership, six countries, no thresholds for PADFA categories.

Decision guidance

Framework mapping and measurement

Metrics that show the programme works:

• Share of active data customers and resellers with a current (under 12 months) foreign adversary control screen.
• Share of live data API keys, storage grants and clean room seats mapped to a screened legal entity (target 100%).
• Share of fields and segments tagged against the 17 categories.
• Median time from a failed screen or ownership change to credential revocation (target same day).

Common failure patterns

• Waiting for a threshold: teams used to Part 202 bulk counts assume small sales are safe. PADFA has no volume minimum.
• Testing at 50%: reusing a sanctions-style 50% ownership rule misses buyers that are 20% to 49% owned from an adversary country.
• Ignoring inferred data: treating modelled interest scores as non-sensitive when they reveal health, religion or military status, and forgetting category (Q).
• Screening the contract, not the credential: the master agreement is screened once, while API keys created later by the customer's affiliates or agents are never tied back to a screened entity.

Practitioner checklist

• Record, per data product, whether it is brokering under 15 U.S.C. 9901(c) and which exclusion applies.
• Tag every field and derived segment against the 17 sensitive data categories.
• Screen all customers and resellers for location, organisation and 20% ownership links to China, Iran, North Korea and Russia.
• Issue data API keys, OAuth clients and storage grants only to screened legal entities, with an owner and an expiry.
• Revoke all machine credentials on a failed screen or ownership change the same day.
• Add foreign adversary warranties, onward transfer bans and ownership change notice to data contracts.
• Screen AI model developers and AI service providers buying training or retrieval data like any other buyer.
• Run PADFA and DOJ Part 202 screening from one counterparty register and check the FTC press releases quarterly.

Standards and references

• Cornell LII: 15 U.S.C. 9901, Prohibition on transfer of personally identifiable sensitive data of United States individuals to foreign adversaries
• FTC: FTC Reminds Data Brokers of Their Obligations to Comply with PADFAA
• FTC: PADFAA warning letter template
• Cornell LII: 10 U.S.C. 4872, covered nations list
• Wiley: FTC Sends Warning Letters to Data Brokers on PADFA Compliance
• Mayer Brown: Cross-Border Transfers of American Personal Information Carry Heightened Regulatory and Litigation Risks

Related NHI Mgmt Group resources: DOJ Data Security Programme Guide · US State Data Broker Laws Compared · California Delete Act and Data Broker Registration Guide · Non-Human Identity (NHI) Guide. This guide summarises the Protecting Americans' Data from Foreign Adversaries Act of 2024 as of 14 November 2026 and is not legal advice.