CFIUS counted health data by what a company keeps

Treasury’s FIRRMA proposal, published in September 2019, put businesses holding health data on more than one million people inside foreign investment review. The count turned on retention and outsourcing, so data inventories became deal and vendor diligence.

Looking Back revisits past weeks one at a time, up to the most recent, and asks what it meant for the banks, health systems and public agencies we serve. The series was written in 2026, with the benefit of hindsight, and each article says which week it looks back on.

Two-minute summaryThe dek and the first sentence of each section, quoted from the article

Treasury’s FIRRMA proposal, published in September 2019, put businesses holding health data on more than one million people inside foreign investment review. The count turned on retention and outsourcing, so data inventories became deal and vendor diligence.

  1. The week On September 24, 2019, the Department of the Treasury’s Office of Investment Security published proposed regulations in the Federal Register, at 84 FR 50174, to implement the Foreign Investment Risk Review Modernization Act of 2018, known as FIRRMA.
  2. The count: one million, measured over time The proposed test was simple to state.
  3. The chain: data held for someone else The definition of a TID U.S. business covered a business that maintains or collects sensitive personal data “directly or indirectly.”
  4. The rights: what an investor gets decides the review CFIUS jurisdiction over a non-controlling investment depended on what the investor received.
  5. Genetic data: no threshold at all The proposal treated genetic information differently from every other category.
  6. Health systems, digital health companies and laboratories The rule applied to businesses, not to sectors, so its effect varied with each organization’s structure and capital.
  7. The strongest objections The argument here is that the proposal made data inventories, and especially retention, a term of investment and of vendor diligence for health care.
  8. With hindsight Treasury published the final rule on January 17, 2020, at 85 FR 3112, effective February 13, 2020.

The week

On September 24, 2019, the Department of the Treasury’s Office of Investment Security published proposed regulations in the Federal Register, at 84 FR 50174, to implement the Foreign Investment Risk Review Modernization Act of 2018, known as FIRRMA. Treasury had released the proposal on September 17. Its press release said FIRRMA had been enacted with overwhelming bipartisan support in August 2018, and that, as the statute required, a final rule would take effect no later than February 13, 2020. The Federal Register notice set a comment deadline of October 24, 2019. The same release included a companion proposal on certain real estate transactions by foreign persons.

The Committee on Foreign Investment in the United States, or CFIUS, is an interagency committee chaired by Treasury. Before FIRRMA, its jurisdiction centered on transactions that could give a foreign person control of a U.S. business. The proposal implemented new jurisdiction over certain non-controlling investments in three kinds of U.S. business: those involved with critical technologies, those performing specified functions for covered critical infrastructure, and those that maintain or collect, directly or indirectly, sensitive personal data of U.S. citizens. The proposal called these “TID U.S. businesses,” for technology, infrastructure and data.

A non-controlling investment in a TID U.S. business would be a “covered investment” if it gave a foreign person, other than an excepted investor, any of three things. The first was access to material nonpublic technical information. The second was membership or observer rights on the board, or the right to nominate a board member. The third was involvement, other than through voting shares, in substantive decisionmaking about the business’s critical technologies, covered infrastructure or, in the data case, the use, development, acquisition, safekeeping or release of sensitive personal data of U.S. citizens.

The definition of sensitive personal data did the work. It covered identifiable data in ten categories, held by a business that met one of three tests. The business targeted or tailored its products or services to U.S. agencies with intelligence, national security or homeland security responsibilities, or to their personnel and contractors. Or it had maintained or collected such data on more than one million individuals at any point over the preceding twelve months. Or it had a demonstrated business objective to do so, where the data was an integrated part of its primary products or services.

Two of the ten categories bore directly on health: the set of data in an application for health, long-term care, professional liability, mortgage or life insurance, and data relating to the physical, mental or psychological health condition of an individual. The others covered financial distress or hardship, consumer reports, non-public electronic communications, geolocation, biometric enrollment data, data used to generate government identification cards, and security clearance information. Separately, the proposal covered all genetic information, as defined in the HIPAA regulations, with no one-million test and no identifiability requirement. Data a business held about its own employees was generally carved out, as was data that is a matter of public record.

The proposal also required a mandatory declaration for a covered transaction in which a foreign person acquired a substantial interest in a TID U.S. business and a foreign government held a substantial interest in that foreign person. Parties could choose to file a full written notice instead.

The September 2019 FIRRMA proposal also covered critical technologies and critical infrastructure. For health care, the important text was the definition of sensitive personal data, and in particular the tests that decided whether a business held enough of it. Those tests did not ask how sensitive one record was. They asked how many people a business held data on, through whom, and with what rights attached to an incoming investor. Each of those questions has a data management answer. The argument here is that the proposal turned the health data inventory, especially how long records are kept and who holds them, into a term of foreign investment and of vendor diligence. The analysis follows six parts: the count, the chain of custody, the investor’s rights, genetic data, how the rule played out for different kinds of health organization, and the strongest objections.

The count: one million, measured over time

The proposed test was simple to state. A business met it if it had maintained or collected data in one of the listed categories on more than one million individuals at any point over the preceding twelve months. The word “maintained” carried as much weight as “collected.” A company that collected health data from a few hundred thousand new customers a year could still cross the line if it kept older records.

Retention sets the number

The final rule, published in January 2020, made this explicit with examples. In one, a company collected physical health data from fewer than one million new customers in the prior twelve months, but under its data retention policy it kept the data for a long time and held it on three million individuals in total. It met the test. In another, the number of records a company held fluctuated under its retention policy. It usually held data on 900,000 people but at one point in the prior twelve months held data on 1,100,000. It met the test as well, even though it held fewer than one million on the day of the transaction.

Those examples describe a mechanism health care compliance officers will recognize. Health records carry retention duties from several directions, and keeping data can be easier than proving it has been deleted everywhere. Under the CFIUS definition, every year of retained data adds to the count. A retention schedule that was written for records management and litigation holds also determined whether a company was a TID U.S. business.

Categories add up

The final rule also changed how the count was framed. It referred to identifiable data within one or more of the listed categories on more than one million individuals. A digital health company that held health condition data on 600,000 people and geolocation data on another 500,000 could not, on that wording, treat the two as separate pools if the people differed. The question became how many distinct individuals appeared anywhere in the covered categories, which is a harder query than counting rows in one database.

The rule also measured the twelve months from a defined point: the earliest of the completion date, certain other transaction events, or the date of a filing. And it let a business show that at completion it would have neither the capability to maintain nor the capability to collect covered data on more than one million people. That exception rewards a company that can prove its data footprint, and it offers nothing to one that cannot.

Identifiable is broader than named

The count applied to identifiable data, and the final rule defined that term broadly. Identifiable data is data that can be used to distinguish or trace an individual’s identity, including through any personal identifier. Aggregated or anonymized data counts as identifiable if any party to the transaction has, or will have as a result of the transaction, the ability to disaggregate or de-anonymize it. Encrypted data is excluded unless the business that holds it has the means to decrypt it.

Each clause maps to a design decision. A de-identified research or analytics dataset stored alongside the linkage table that re-identifies it does not escape the count. Neither does a pseudonymized dataset whose keys the company holds. The investor’s own capabilities matter too: if an incoming investor could re-identify a dataset by combining it with data it already holds, the dataset is identifiable for this purpose. On the other side, a vendor that stores data encrypted under keys held only by its customer has a credible argument that it cannot decrypt that data. Key custody, often treated as a security detail, becomes a fact in the jurisdictional analysis.

For a health system, the practical step is to record, for each dataset described as de-identified, how it was de-identified, who holds any re-identification key, and whether the data has been combined with other sources. That record serves HIPAA de-identification decisions as well as any CFIUS question.

What a deal team has to produce

For an investor and a target, the count is a factual question that has to be answered before signing. That means an inventory, not an estimate. The target needs to list each system that holds identifiable data in a covered category, the number of distinct individuals in each, the overlap between systems, and the peak count over the prior twelve months. It needs the retention schedule that drives those numbers and evidence that the schedule is followed. Backups, archives, analytics copies and test environments all hold data, and each counts as maintaining it.

A privacy program that maps data flows by processing purpose serves notices and contracts, but it does not produce a head count across systems. A security program that catalogs systems by criticality does not either, because criticality is not the number of people whose data a system holds. The CFIUS test asks for a third view: a count of distinct people per category, over time, across every copy. Building it the first time, under deal pressure, can be slow and expensive. Building it as part of ordinary data governance is cheaper and pays off in breach response and records management too.

Deletion as a deal lever

The retention examples also create an option. A company near the threshold could reduce its count by enforcing its own retention schedule and deleting data it no longer needs, so long as it does so before the relevant twelve-month window and in line with its legal obligations. That is not a loophole. It aligns the CFIUS outcome with what privacy and security programs already recommend. But it only works for a company that knows where its data is, and it cannot be done hastily on the eve of a transaction without creating records-retention and legal hold risk. Counsel has to be involved in any such step.

The chain: data held for someone else

The definition of a TID U.S. business covered a business that maintains or collects sensitive personal data “directly or indirectly.” That phrase reaches across contracts. A health care organization may not hold all of its data on its own servers. Claims processors, cloud hosts, revenue cycle firms, transcription services, analytics vendors and patient engagement platforms hold copies.

Both ends of an outsourcing contract

The final rule’s example 12 made the point with a medical device maker. The company solicited patient medical information on its five million customers for research and development, marketing and quality assurance. It did not hold the data itself. A third party collected it on the company’s instructions and maintained it on the third party’s servers for the company to access. The rule said both were TID U.S. businesses: the device maker because it indirectly maintained and collected the data, and the vendor because it did so directly.

That symmetry matters for health care procurement. A vendor that hosts or processes identifiable health data for several clients may hold data on far more than one million people in aggregate, even if no single client is that large. A foreign investment in that vendor, with a board seat or with involvement in decisions about how the data is used or safeguarded, would be a covered investment in a TID U.S. business. The clients would not be parties to that transaction. Their patients’ records would still be the reason it mattered.

What the customer can see

The proposal gave a health system no role in a vendor’s CFIUS review. CFIUS reviews the parties to a transaction. A hospital that contracted with a vendor in 2019 had no regulatory channel to learn that a foreign investor was taking a board seat at the vendor, or to see any mitigation terms CFIUS imposed. Its only levers were contractual.

Those levers are familiar. A contract can require notice of a change of control or of a material investment, define those terms to include board or observer rights, and give the customer the right to review and, in defined cases, to terminate. It can require the vendor to state whether it has filed with or been reviewed by CFIUS in connection with a transaction, to the extent the vendor may disclose that. It can require that any investor access to customer data follow the same controls as any other access, logged and limited to need. Data processing terms address how data is used and may say nothing about who owns the vendor. Ownership and governance changes belong in the master agreement.

Inventory once, use twice

The same inventory that answers the CFIUS count also tells a health system which vendors hold its patients’ data and how many records each holds. A system that keeps that register current can answer both a deal question, when it is the target or a partner, and a procurement question, when it is the customer. Third-party risk programs that classify vendors only by service type and contract value miss the dimension the CFIUS rule used: the number of identifiable people whose data the vendor keeps.

What the vendor register should hold

A register that supports both uses needs a handful of fields for each vendor that holds identifiable patient data:

  • The categories of data held, mapped to the categories in the rule, such as health condition data, insurance application data and genetic test results.
  • The number of distinct individuals whose data the vendor holds for the organization, and whether the vendor holds the data or only processes it in transit.
  • The vendor’s retention period for that data and the evidence that deletion happens at the end of it.
  • Who holds the encryption keys, and whether the vendor can decrypt the data.
  • The vendor’s ultimate ownership as last confirmed, the date of confirmation, and any known investor with board, observer or data decisionmaking rights.
  • The contract clauses that require notice of ownership or governance changes, and the exit terms if those change.

Many of these fields may already exist somewhere: in security questionnaires, contracts and privacy assessments. Pulling them into one record per vendor is the work. Updating ownership at each contract renewal, and on notice of any financing, keeps it current.

The rights: what an investor gets decides the review

CFIUS jurisdiction over a non-controlling investment depended on what the investor received. A passive minority stake with no board role, no access to material nonpublic technical information and no involvement in substantive decisions about sensitive personal data was outside the new jurisdiction. Add one of those rights and the same investment could be covered.

Rights are drafted, not discovered

That makes the covered-investment question a drafting question. Investment agreements, side letters and governance documents set board seats, observer rights, information rights and consent rights. A clause that gives an investor a veto over a data-sharing partnership, or a seat on a committee that approves analytics uses of patient data, is the kind of involvement in decisionmaking the proposal named. A clause that gives the investor quarterly financial statements and nothing more is not.

Investment funds got their own treatment. The proposal excepted from the mandatory declaration an investment through a fund managed exclusively by a general partner who is not a foreign person, where any foreign limited partner sitting on a fund advisory board or committee met specified conditions. The final rule spelled those conditions out: among them, the advisory board could not approve, disapprove or otherwise control the fund’s investment decisions, and the foreign person could not otherwise control the fund. For health care companies raising money through funds with foreign limited partners, the structure of the fund and the rights of those partners became part of the analysis.

The data governance terms in a term sheet

For a digital health company, the practical result was that data governance terms belonged in the term sheet. A company could accept foreign capital while keeping the investor away from decisions about the use, safekeeping or release of sensitive personal data. It could also accept the rights and the review. Either path was legitimate. What the proposal penalized was not knowing which path the documents described.

Excepted investors

The new jurisdiction did not apply to every foreign investor. FIRRMA required the regulations to limit it to certain categories of foreign persons, and the proposal did this through “excepted foreign states” and “excepted investors.” In the final rule, the Committee initially selected Australia, Canada and the United Kingdom as eligible foreign states, citing their intelligence-sharing and defense industrial base integration with the United States. An investor that met the excepted investor criteria fell outside the non-controlling investment jurisdiction, though transactions that could result in control remained reviewable.

For a health care company, the investor’s home country and corporate structure therefore became part of the term sheet analysis. The same board seat could put one investment inside CFIUS’s jurisdiction and leave another outside it.

That has an operational consequence. If a company promises an investor no access to identifiable data, its access controls have to make the promise true. An investor’s representative on the board who receives patient-level analytics in a board pack, or who can log into a reporting tool, has access whatever the term sheet says. The board materials process, the reporting platform’s permissions and the data room for the transaction itself all need to match the rights the documents grant.

The calendar a mandatory declaration sets

Where the mandatory declaration applied, the proposal required it no later than thirty days before the completion date of the transaction, with transition rules for deals closing soon after the final rule took effect. That puts the CFIUS analysis before signing, not after. A buyer or investor needs the target’s data count, the rights in the draft documents and the ownership of its own investors early enough to decide whether to file, and to choose between a short declaration and a full written notice.

For a health system partnering with a foreign-backed company in a joint venture, the same calendar applies to the venture. The system’s legal, compliance and data teams have to supply facts about the data the venture will hold, on a deal timetable that may be set without them.

Joint ventures and partnerships

The rule defined a transaction to include the formation of a joint venture. Health systems may form joint ventures for imaging centers, ambulatory surgery, laboratory services, specialty pharmacy and data platforms. Where a foreign partner joins a venture that maintains or collects identifiable health data on more than one million people, the CFIUS questions apply to the venture. A health system’s strategy team may see the deal as an operating partnership. The definition sees a foreign person acquiring rights in a TID U.S. business. The venture agreement’s governance and data clauses, read against the rule, decide which view is right.

Genetic data: no threshold at all

The proposal treated genetic information differently from every other category. It did not require that the business meet the one-million test or target sensitive government populations, and it did not require that the data be identifiable. Any genetic information, as defined in the HIPAA regulations, counted.

Any amount, any form

The effect was that any company holding genetic information, at any scale, was a TID U.S. business for purposes of the proposal. That reached laboratories, genetic testing companies, biobanks, research platforms and health systems that hold genetic test results in their records.

How the final rule narrowed it

Commenters pushed back. Some asked to limit the definition to identifiable data or to full genomes. Others asked to exclude anonymized data from drug discovery or clinical trials, or aggregated data from large heterogeneous populations. The final rule recalibrated the provision in two ways. It focused on the results of genetic tests, using the definition of “genetic test” from the Genetic Information Nondiscrimination Act of 2008, and it limited coverage to identifiable data. It also carved out genetic testing data derived from databases maintained by the U.S. Government and routinely provided to private parties for research.

The narrowing left a sharp line where none had existed before. Identifiable genetic test results, including related sequencing data, counted with no threshold. De-identified research datasets generally fell outside the genetic category. For a research institution or a laboratory, the identifiability of each genetic dataset became a fact with legal consequences in an investment, and the de-identification method behind it became something counsel would want documented.

Genetic results inside clinical records

Genetic test results do not live only in laboratories and research platforms. When a health system orders genetic testing for patients, the results become part of their records. On the final rule’s text, a business that maintains identifiable genetic test results is in the genetic category at any volume. A health system’s laboratory outreach business, a precision medicine program spun into a separate company, or a data platform that includes identifiable genetic results would each carry that status into any transaction involving foreign capital.

That makes a precise inventory of where identifiable genetic results are stored more valuable than a general statement that the organization holds genetic data. If a planned transaction concerns only part of the organization, the question is whether identifiable genetic results sit in that part.

Proving a genetic dataset is not identifiable

The two definitions work together. A genetic dataset falls in the category only if it is identifiable, and the definition of identifiable data reaches aggregated or anonymized data whenever any party to the transaction can disaggregate or de-anonymize it. A research institution claiming that a dataset is outside the category therefore has to show more than the removal of names. It has to show what was removed, what linkage keys exist and who holds them, and that neither the company nor the incoming investor can re-link the records. Those facts are technical, and they change when datasets are combined. Writing them down when the dataset is created is likely to be far cheaper than reconstructing them during diligence.

Research collaborations

Academic medical centers may share genetic data with outside collaborators under data use agreements. A collaboration is not an investment, and the CFIUS rule did not reach it as such. But a research platform company that pools identifiable results from several institutions, and then takes foreign investment, sits squarely inside the genetic category. The institutions supplying data have the same interest as any vendor customer: to know who will own and govern the platform, and to keep the contractual right to stop supplying data if that changes.

Health systems, digital health companies and laboratories

The rule applied to businesses, not to sectors, so its effect varied with each organization’s structure and capital. Precision matters here because the obligations sit with the parties to a covered transaction.

Nonprofit health systems

A nonprofit health system has no shareholders and issues no equity. For such a system, the direct effect of the rule was narrow: joint ventures with foreign partners, minority stakes in for-profit subsidiaries, and investments by a health system’s venture arm in companies that later take foreign capital. The indirect effect was larger. Some of their vendors and partners were venture-backed companies with mixed investor bases, and the health system’s patient data was part of what made those companies TID U.S. businesses.

Investor-owned providers and physician groups

For-profit hospital operators, physician practice platforms and ambulatory networks backed by private equity are likely to be more exposed. Their funds may include foreign limited partners, and their sale processes may include foreign bidders. A large ambulatory network can hold identifiable health data on more than one million people, especially once years of retained records are counted. For these organizations, the CFIUS analysis belongs in the sale and financing process alongside antitrust and licensing approvals.

Digital health and health data companies

On our reading, venture-funded digital health companies were among the most directly affected. A company with a demonstrated business objective to collect covered data on more than one million people met the test before it reached that number, provided the data was an integrated part of its primary products or services. If it also raised money from investors outside the United States, every round carried the question. The final rule recorded that some commenters argued the provisions could hinder the growth of financial technology start-ups, and the same pressure applied to health technology. For these companies, the CFIUS position became a recurring diligence item at every financing round.

Laboratories and genetic testing companies

Under the final rule, a laboratory that holds identifiable genetic test results was in the genetic category regardless of volume. Clinical laboratories owned by health systems, independent reference laboratories and consumer genetic testing companies all fit. Their owners and investors had to treat identifiable genetic results as the most sensitive asset in any transaction.

Device and remote monitoring companies

The final rule’s outsourcing example was a medical device maker, and the choice fits the sector. Connected devices and remote monitoring services can collect health condition data continuously, and the geolocation category in the proposal expressly named data collected through wearable electronic devices and mobile applications. A company whose product combines both can reach the count through either category, and the rule’s wording, which covers data within one or more categories, points to counting distinct individuals across categories. Where the company relies on a cloud or analytics vendor to hold the data, the example says both the company and the vendor are TID U.S. businesses.

Health systems that prescribe or resell such devices sit one step further back. Their patients’ data may be held by the device company, not the system. The vendor register described above is the system’s only view of how many of its patients’ records the company keeps, and of who owns it.

Who does what inside a health system

The work divides cleanly. The general counsel’s office adds a CFIUS screen to the intake for joint ventures, minority investments and the venture arm’s portfolio. The privacy and security teams own the data inventory and the identifiability records. Procurement owns the vendor ownership clauses and the register fields that track them. The venture or strategy team asks each portfolio company and partner for its CFIUS position at each financing.

Two measures show whether the work is done. One is the share of vendors holding identifiable patient data whose record includes a current count of individuals and a confirmed ownership date. The other is how long it takes to produce, on request, the count of distinct individuals per covered category across the organization. If the answer is weeks, the inventory is not ready for a transaction.

Payers and insurers

The category for applications for health and life insurance reached health plans and insurers directly. An insurer holding application data on more than one million people met the test on that category alone. For such an insurer, the rule added a federal national security review for non-controlling foreign investments that carry board or decisionmaking rights.

The strongest objections

The argument here is that the proposal made data inventories, and especially retention, a term of investment and of vendor diligence for health care. Three objections deserve an answer.

CFIUS reviews few health care deals

Many health care transactions involve no foreign person and never reach CFIUS. The argument does not depend on frequency. The rule changed what a well-run company needs to know about itself. A company that cannot say how many people it holds data on, across which systems and for how long, cannot answer the CFIUS question when it arises. It also cannot answer a breach notification question, a records request or a regulator’s inquiry efficiently. The CFIUS test is one more reason to have an inventory that should already exist.

Ownership was the wrong place to look

The strongest objection comes from the other direction. It says the proposal looked at the wrong channel. A foreign investor with a board seat is visible, documented and, under the rule, reviewable. A vendor whose staff or subcontractors abroad can reach patient data through a support connection is none of those things, and the CFIUS rule did not reach it at all, because a service contract is not an investment. On this view, a health system worried about foreign access to patient data should spend its effort on vendor access paths, not on the cap tables of its suppliers.

The objection is largely right about where the larger exposure lies, and the federal government later moved in that direction, as the hindsight section describes. It does not defeat the argument here. The inventory that CFIUS made necessary, of how many people’s data sits where and with whom, is the same inventory a review of vendor access paths needs. Ownership diligence and access diligence draw on one register. An organization that built it for one purpose had most of what it needed for the other.

The count is a blunt instrument

A second objection is that one million is arbitrary and that counting people is a poor proxy for national security risk. Commenters said so. Some suggested five million U.S. citizens instead, arguing the lower threshold would capture too many businesses that pose no risk. The final rule kept one million. Whatever its merits as policy, a bright line has one advantage for operations: it can be tested. A qualitative standard would have left companies guessing. A number tells them exactly what to count.

Counting costs small companies most

A related objection is cost. A young digital health company may not have a data engineer to spare for an inventory that its customers never asked for. The concern is fair, and it argues for proportion: a small company with one product and one database may be able to produce a credible count in days. The cost rises with the number of systems, copies and partners, which is to say with the very complexity that makes a count hard to verify. A company that defers the work until a term sheet arrives pays for it at the worst moment.

Contracts cannot fix ownership risk

A third objection is that a health system’s contract terms cannot prevent a foreign investor from influencing a vendor. That is partly right. Contracts give notice and exit rights, not control. But exit rights matter in procurement. A health system that learns of a change in a vendor’s ownership and governance can reassess the vendor, require additional controls, or plan a transition. One that learns nothing has no options. The contract is the only instrument a customer has, so it should be used.

Where the argument is weakest

The weakest point is visibility. The proposal created duties for the parties to a transaction. A health system is not a party to its vendor’s financing and has no right under the rule to learn whether the deal was filed or what mitigation applied. The argument holds as a matter of governance: know your own count, know your vendors’ ownership, and write contracts that tell you when either changes. It does not promise that those steps will reveal every risk.

With hindsight

Treasury published the final rule on January 17, 2020, at 85 FR 3112, effective February 13, 2020. It kept the one-million threshold, despite comments asking for five million, and recorded commenters’ concern that the threshold might capture companies posing no national security risk. It measured the twelve months from the earliest of completion, certain other events, or a filing, and let a business show it lacked the capability to hold covered data on more than one million people. It narrowed genetic data to identifiable genetic test results, excluding certain U.S. Government research databases. It initially selected Australia, Canada and the United Kingdom as excepted foreign states. Its examples on retention and outsourcing are described above. The definition of sensitive personal data in the Code of Federal Regulations today reads as the final rule did.

CFIUS later raised the stakes for errors. A final rule published November 26, 2024, effective December 26, 2024, increased the maximum civil penalty for a declaration or notice with a material misstatement or omission, or a false certification, from $250,000 to $5,000,000 per violation, and raised the maximum for violating a mitigation agreement entered into on or after December 26, 2024, to the greatest of $5,000,000, the value of the violator’s interest in the U.S. business, or the value of the transaction filed with the Committee.

The bigger change came from outside CFIUS. Executive Order 14117 of February 28, 2024, led to a Department of Justice final rule, published January 8, 2025, at 90 FR 1636, and effective April 8, 2025. It prohibits or restricts certain data transactions with six countries of concern (China, Cuba, Iran, North Korea, Russia and Venezuela) and with covered persons. Where CFIUS looked at who invests in a business, the Justice Department rule looks at who gets access to bulk data through data brokerage, vendor, employment or investment agreements. Its thresholds are much lower: personal health data on more than 10,000 U.S. persons, and human genomic data on more than 100. One of its examples is a U.S. medical facility that hires an IT services company headquartered in a country of concern with access to systems holding bulk patient health data. That vendor agreement is a restricted transaction. Investment agreements subject to a CFIUS action are exempt from most of the rule. Due diligence, audit and certain reporting duties for restricted transactions applied to transactions on or after October 6, 2025. The country list in the Code of Federal Regulations today is unchanged.

In 2026, Treasury turned to speed. On February 9, 2026, Treasury asked for comment on a Known Investor Program and on streamlining CFIUS reviews, citing the February 2025 “America First Investment Policy” memorandum. It said about 70 percent of covered transactions over the past five years were approved in the initial review phase, and more than 90 percent were approved overall. Comments were due March 18, 2026.

What held up was the 2019 premise that a count of identifiable people is a security attribute. What changed was that the contract came under federal review alongside the investor. By 2025 the federal government was asking a similar question of vendor contracts, for health data, at 10,000 people rather than one million.

Spotted an error? Tell us and we will check it and correct the article where needed.

  • CFIUS
  • FIRRMA
  • sensitive personal data
  • health data
  • vendor risk
  • genetic data