A ban on anything Kaspersky developed “in whole or in part” pointed agencies inside their products
Section 1634 of the defense authorization act, signed December 12, 2017, barred federal use from October 1, 2018 of anything a Russian security vendor developed or provided, even in part. DHS’s directive had left embedded code out. In our reading, removal became a component inventory.
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
Section 1634 of the defense authorization act, signed December 12, 2017, barred federal use from October 1, 2018 of anything a Russian security vendor developed or provided, even in part. DHS’s directive had left embedded code out. In our reading, removal became a component inventory.
- The week On Tuesday, December 12, 2017, the President signed H.R. 2810, the National Defense Authorization Act for Fiscal Year 2018.
- Scope: a brand in the directive, an origin in the statute The two instruments overlapped.
- Discovery: finding a vendor inside other vendors’ products The directive’s first demand was an inventory.
- Replacement: removing a security tool removed a capability The directive’s plan template asked a question that turns removal into an architecture task.
- Supply chain: the duty moved to contractors The statute bound federal elements.
- How the ban landed on different public bodies The statute applied to the Federal Government.
- Objections, in order Five objections deserve an answer.
- What a leader could check These checks follow from the four parts.
- With hindsight The district court that later heard the vendor’s challenges filled in the December timeline.
The week
On Tuesday, December 12, 2017, the President signed H.R. 2810, the National Defense Authorization Act for Fiscal Year 2018. It became Public Law 115-91. Among its hundreds of sections was section 1634, a prohibition on federal use of products and services developed or provided by Kaspersky Lab, a security software maker headquartered in Russia.
Subsection (a) set the rule. No department, agency, organization or other element of the Federal Government could use any hardware, software or services developed or provided, in whole or in part, by the company or any successor. The ban also reached any entity that controls, is controlled by or is under common control with it, and any entity it majority-owns. It covered use directly or through work with or on behalf of another federal element. Subsection (b) made the ban take effect on October 1, 2018.
Subsection (c) asked for a wider review. The Secretary of Defense was to consult five other officials, including the Secretary of Homeland Security and the Director of National Intelligence. Together they were to review the procedures for removing suspect products or services from federal networks. The report was due 180 days after enactment. It was to describe government-wide exclusion authorities, gaps in them, monitoring methods, information sharing and recommendations.
Agencies were already working under an earlier instrument. On September 13, 2017, the Acting Secretary of Homeland Security had issued Binding Operational Directive 17-01 on the same vendor’s products. A binding operational directive is a compulsory direction to federal executive branch agencies under the Federal Information Security Modernization Act of 2014. It does not apply to national security systems or to certain Defense Department and intelligence community systems.
The directive set three deadlines. Within 30 days, agencies had to identify the covered products on their federal information systems and report to DHS. Within 60 days, they had to deliver a detailed removal plan. At 90 days, unless DHS directed otherwise based on new information, they had to begin implementing the plan and report progress every 30 days until removal was complete.
The directive’s definition was the brand. It covered information security products, solutions and services supplied directly or indirectly by the company or its predecessors, successors, parents, subsidiaries or affiliates. It listed ten products by name, from consumer antivirus to an embedded systems security product. It said in terms that it did not address the company’s code embedded in the products of other companies. It also left out two services, threat intelligence and security training.
The removal plans had a long list of required elements. They included the products found, the number of endpoints, the methods used to find them, the mission functions of affected systems and every contract or service agreement with the vendor. They also covered any security controls removal would affect, replacement products, timelines and costs. DHS noted possible budget implications and told CIOs and procurement officers to coordinate with their chief financial officers.
DHS had also set up a process for objections. Its Federal Register notice of September 19, 2017, gave the company and other directly affected entities until November 3 to respond. It said the Secretary’s decision would be communicated in writing by December 13, 2017, the Wednesday of the week the statute was signed. According to the district court that later heard the company’s challenge, the Acting Secretary issued a final decision on December 6, 2017, and maintained the directive without modification. So in the week of December 11, agencies held a directive that had just been confirmed and had entered its removal phase, and a new statute with a wider reach.
A ban on a named vendor sounds like the simplest instruction in information security. Find the product, uninstall it, buy something else. The two federal instruments on Kaspersky Lab in late 2017 show why that picture was incomplete. They used different definitions, and the difference decided how much work an agency faced.
The DHS directive defined its target by brand and said it did not reach Kaspersky code embedded in other companies’ products. The statute signed on December 12 defined its target by origin: anything developed or provided, in whole or in part, by Kaspersky Lab or related entities. In our reading, that phrase moved compliance from the endpoint console to the component level. An agency could no longer satisfy the rule by searching for a product name. It had to know what was inside the products it ran.
Our argument is that section 1634 made vendor removal an architecture problem. The relevant questions became which licensed engines, libraries and services sat inside an agency’s tools, and who could answer that question. That is a different capability from scanning endpoints for a brand. In our view, few agencies had it as a routine function in 2017.
The difference was not a drafting accident. The directive was written by an agency acting under FISMA, which gives DHS authority over federal information systems in the executive branch. It had to rest on a determination about a known or reasonably suspected risk, and it named the products it had assessed. The statute was written by Congress, which in our reading needed no such assessment and could define its target as widely as it chose. It chose the vendor’s involvement in a product, whatever the product was called.
For agency technology leaders, the result was two rules on the same vendor, a few months apart, with different scopes and dates. The first could be met with standard endpoint tools. The second, as we see it, assumed knowledge of supply chains that agencies had seldom been asked to produce.
We take the problem in four parts. Together they cover what the rules reached, how an agency could find it, what removing it broke, and how the duty passed to suppliers:
- Scope: what the directive and the statute each covered, and where they differed.
- Discovery: how an agency could find a vendor’s products and a vendor’s code inside other products.
- Replacement: what removing a security product took away, and how to replace it without a gap.
- Supply chain: how the rule passed from agencies to the contractors who sell to them.
After the four parts we look at how the ban landed on different public bodies, take the main objections in order, and list what a leader could check. A note on our reading: the statute’s words are quoted from the public law, and the directive’s from DHS. Where we say what the words required in a particular case, that is our reading, and agencies and contractors should check with counsel.
Scope: a brand in the directive, an origin in the statute
The two instruments overlapped. Each reached Kaspersky products on federal systems. But they differed in four ways that mattered.
What was covered
The directive covered information security products, solutions and services supplied directly or indirectly by Kaspersky and its affiliates. It named ten. It excluded code embedded in other companies’ products and two named services.
The statute covered any hardware, software or services developed or provided, in whole or in part, by Kaspersky Lab, its successors, entities in its control group and entities it majority-owned. It was not limited to security products. It contained no exclusion for embedded code or for particular services.
Whom it bound
The directive bound federal executive branch agencies through FISMA. By statute, it did not apply to national security systems or to certain Defense Department and intelligence community systems.
The statute bound every department, agency, organization or other element of the Federal Government. It contained no carve-out for national security systems. In our reading, that brought systems outside DHS’s directive authority within the ban, including Defense and intelligence systems.
How it reached use through others
The statute covered use directly or through work with or on behalf of another federal element. In our reading, that language reached shared services. An agency that relied on another agency’s hosted service would be within the ban if that service used a covered product.
When it bit
The directive’s clock was already running: identification by mid-October, plans by mid-November and removal starting on December 12, 2017. The statute’s ban took effect on October 1, 2018. So in the week the statute was signed, agencies were starting to remove branded products under one instrument while learning that a wider rule would follow under another.
Where the directive went further
The statute was wider in scope, but the directive was more detailed in process. It required reports to DHS, a plan with fourteen listed elements, and status reports until removal was complete. DHS said it would rely on agency self-reporting and independent validation measures to track progress. The statute required none of that from agencies. It stated a prohibition and a date. In our reading, the directive supplied the management machinery and the statute supplied the reach, and an agency needed both.
Two clocks in one week
The week of December 11 put both clocks in front of agencies together. The directive’s removal phase began on December 12, with status reports to follow every 30 days. The statute, signed the same day, set a later effective date with a wider reach. An agency could reasonably have planned one removal program to satisfy both: remove branded products now, and use the months before October 2018 to find embedded components. In our reading, treating them as two separate projects would have meant doing discovery twice.
Why the difference in definitions mattered
A brand-based rule can be checked against a product list. An origin-based rule requires knowing the provenance of components. Security software can be built into other companies’ products. The 2024 Commerce determination, discussed below, named that practice as the integration of Kaspersky software into third-party products and white labeling, and said the user would be less likely to know the true source of the code.
In our reading, the statute’s in-whole-or-in-part phrase reached that case. A firewall, gateway or email filter whose scanning engine came from Kaspersky Lab would be hardware, software or services developed in part by Kaspersky Lab. The directive had expressly excluded it.
What the statute did not say
The statute was short, and it left questions open. It did not define use. It did not say whether a component that was present but disabled counted. It did not define developed, so it did not say whether code written years earlier and since rewritten by another company was still developed in part by the original author. It did not provide a waiver process or exceptions for research or threat analysis.
Each gap left agencies to choose a reading. A cautious reading treated any presence of covered code as use. A narrower reading looked at whether the component was active. In our reading, the absence of a waiver made the cautious reading the safer one for an agency, because there was no route to approval for a borderline case. The acquisition rule later answered some of these questions for contractors, but not all of them.
Discovery: finding a vendor inside other vendors’ products
The directive’s first demand was an inventory. Agencies had to report the products found, the number of endpoints affected and the methods they used. Asking for the method is telling. DHS wanted to know how agencies looked as well as what they found.
Finding branded products
For branded products, discovery used familiar tools. Endpoint management and software asset inventories list installed applications by publisher and product name. Network scans can find agent traffic to known update servers. Procurement records show licenses bought.
Each source has blind spots. Inventories miss unmanaged devices and systems outside the agency’s main management domain. Procurement records miss software bought on a purchase card or bundled with something else. Network monitoring misses systems that do not call home. The directive’s request for methods let DHS judge coverage.
Who owned the question
Discovery crossed organizational lines. Asset management owned the endpoint inventory. The security operations team knew what tools were deployed. Procurement held the contracts and could write to suppliers. Program offices knew what their contractors ran. No one of them could answer the statute’s question alone. An agency that named a single owner for the component inventory, with authority to ask the others, was better placed to answer it.
Systems outside central management
We would expect most agencies to have systems their central tools do not see well. Laboratory instruments, building control systems, field office equipment and mobile devices can run their own security software or none. Systems run by contractors on the agency’s behalf are federal information systems under the directive’s definition, but the agency does not administer them directly. Each of these needed its own discovery method, from a site survey to a written question to the contractor.
Finding embedded code
Embedded code is harder. A product that licenses another company’s engine does not, by default, show that engine in an installed-software list. The evidence lives in the product’s documentation, its license notices, its update traffic or the vendor’s own statement.
So discovery under the statute depended on asking suppliers. An agency could write to each vendor of security products, network appliances and managed services and ask whether any component was developed or provided by Kaspersky Lab or a related entity. The answer had to come from the supplier, because the agency could not see inside the product.
The limits of attestation
Asking suppliers works only as well as suppliers know their own products. A large vendor can trace its licensed components. A small reseller of a niche appliance may not know whose engine is inside. A supplier may also rely on its own suppliers, so the answer passes through several hands before it reaches the agency.
An agency can strengthen attestations in three ways. It can ask specific questions, naming the vendor and its known products, rather than general ones. It can ask for the basis of the answer, such as a list of licensed components. And it can compare answers against its own evidence, such as observed update traffic or license notices in product documentation. The result is still not certainty. It is better evidence than a signed form alone.
Services and hosted platforms
Services added another layer. A managed security provider might use a Kaspersky engine in its own infrastructure. A cloud email service might scan attachments with a licensed engine. The agency did not install those components. Under the statute’s use-through-others language and its coverage of services, they still counted in our reading.
Subscriptions and information services
The directive excluded two of the vendor’s services, threat intelligence and security training. The statute had no such exclusion. It covered services developed or provided in whole or in part by the company. In our reading, an agency that subscribed to the vendor’s threat feeds or used its training content would have to consider them under the statute, even though the directive had left them alone. Subscriptions are easy to miss in a software inventory, because nothing is installed. They show up in purchase records, in security team tool lists and in the feeds configured in other products, such as a security information and event management system.
What an inventory needed to record
To meet the statute, an inventory would need more than product names. For each security and network product, it would record the vendor, the version, the components and licensed engines inside it, and the source of that information. For each service, it would record what tools the provider used to deliver it. That is a component inventory, the kind later discussions would call a software bill of materials.
We do not claim agencies built such inventories in response. The point is that the statute’s wording assumed a capability that a brand-based inventory could not supply.
Evidence an inspector general would ask for
An auditor testing compliance with an origin-based ban would want three things. First, the agency’s method for identifying covered articles, including how it reached embedded components. Second, the supplier attestations or other evidence it relied on. Third, the record of any covered article found and what was done about it. An agency that kept only a list of uninstalled products would struggle to show the second and third.
Keeping the inventory current
An inventory taken once in late 2017 would not stay true. Products update. Vendors change licensed components between releases. Agencies buy new tools. An origin-based ban that runs indefinitely calls for an inventory that is kept current: new purchases checked at intake, updates reviewed when a supplier changes its components, and periodic reconciliation. That is ordinary asset management discipline, applied to a question asset management did not routinely ask.
Replacement: removing a security tool removed a capability
The directive’s plan template asked a question that turns removal into an architecture task. If removing the product would affect FISMA performance requirements or security controls, the plan had to say which.
What the directive’s list revealed
The list DHS gave was long. It included data loss prevention, network access control, mobile device management, sandboxing, web reputation and content filtering, whitelisting of hardware and software, vulnerability and patch management, anti-malware, anti-exploit, spam filtering and encryption. Security suites provide several of these at once. Removing one suite could open gaps in several controls.
Sequencing without a gap
The safe order is replacement before removal. Choose the replacement, test it, deploy it alongside, verify coverage, then remove the old product. The directive’s plan asked for the replacement product and its timeline as well as the removal timeline. That ordering assumes money and time.
Removal before replacement leaves systems without protection. For endpoint security, that gap is a real exposure. The directive let agencies set their own removal timeline within the plan, which in our reading allowed for sequencing.
Choosing the replacement
The replacement had to pass the same test as everything else. A new security product that licensed a covered engine would itself be a covered article under the statute’s wording. So the selection process needed the component question built in. In our reading, the safest sequence was to ask each candidate supplier for its component disclosure before evaluation, not after award.
Users and the help desk
Endpoint security changes reach every user. A new agent can slow machines, block tools users rely on or trigger alerts that the old product did not. Help desks need scripts and escalation paths before the rollout, and users need notice. For agencies with remote and field staff, the rollout also depends on devices connecting to the network long enough to receive the new agent and remove the old one.
Costs and the budget office
DHS anticipated costs. The plan had to estimate license, maintenance and replacement costs, coordinated with the chief financial officer. The directive’s section on budgetary implications said compliance could have budget effects and told CIOs and procurement officers to coordinate with the CFO.
A replacement bought in the middle of a fiscal year needs money from somewhere. That may mean reprogramming funds or deferring other work. For a small agency with one security suite across its whole estate, the cost could be large relative to its IT budget.
Configuration and data
Replacing a security product is also a data migration. Policies, exclusions, alert rules and historical detection logs live in the old product’s console. Some can be exported. Some must be rebuilt. Logs that support investigations or audit trails need to be retained under the agency’s records schedules even after the tool that made them is gone.
Proving removal
A removal plan ends with evidence. The directive asked for status reports every 30 days until full removal. To show removal, an agency needed a before list, an after list and a method for confirming that uninstalled agents did not return through images, backups or reinstallation scripts. We think golden images and deployment packages are a likely place for removed software to persist, because they are built once and reused.
Decommissioning also covers licenses and accounts. Management consoles, cloud tenants and vendor portal accounts should be closed and their credentials retired. Any data the product sent to the vendor’s infrastructure stays there, out of the agency’s reach. Removal stops future flows. It does not recall past ones.
Embedded components are harder to replace
For a branded product, replacement is a purchase. For an embedded component, the agency depends on the supplier. The supplier may swap the engine in a later release, or the agency may have to replace the whole appliance. That decision is outside the agency’s control and on the supplier’s product roadmap.
Supply chain: the duty moved to contractors
The statute bound federal elements. But agencies buy almost everything they run. The practical way to enforce an origin-based ban is through contracts, so that suppliers certify and report. The supplier is the party that knows what is inside its product. A contract can require it to say so, to warrant that the product is clean and to report when it learns otherwise. That moves the information to where the agency can use it, and gives the agency a remedy if the supplier is wrong.
The acquisition rule that followed
The FAR Council did that in 2018, as described in the hindsight section. Its clause defined a covered article as any hardware, software or service that is developed or provided by a covered entity, includes such a component, or contains components using such hardware or software. That definition spelled out the embedded case the statute’s phrase implied.
The clause also reached contractors’ own work. It prohibited using covered articles in the development of data or deliverables first produced under the contract. So a contractor’s internal tools mattered as well as what it delivered.
Reporting on a short clock
The clause required contractors to report covered articles they found, or learned about from subcontractors, in writing. Under the 2018 interim clause, the first report was due within one business day, with contract numbers, supplier, brand, model and any mitigation. The current version of the clause, dated December 2023, allows three business days for that first report. A fuller report followed within ten business days. The clause flowed down to subcontracts.
That clock assumes a contractor can find out quickly what is inside its products. A reseller of a third-party appliance would depend on the manufacturer’s answer. In our reading, the clause made supplier transparency about components a contract term.
Orders under larger contracts
The clause anticipated the way agencies buy technology through indefinite delivery contracts. For those, the contractor had to report to the contracting officer for the base contract and to the contracting officers for any affected orders. Defense Department contractors reported through a designated DoD website instead. That detail matters to an agency because one report can touch orders placed by several offices. Someone has to make sure each affected office hears about it.
Contractors’ own environments
The clause’s second prohibition reached inward. Contractors could not use covered articles in the development of data or deliverables first produced under the contract. That brought the contractor’s own laptops, build servers and security tools into scope. A software developer delivering code to an agency would need to confirm that its development environment did not run covered products. Its internal IT team became part of its federal compliance program.
Resellers and distributors
Federal buyers can purchase through resellers, distributors and contract vehicles rather than from manufacturers. The clause’s reporting duty fell on the prime contractor, with flowdown to subcontractors. A reseller holding a government contract would have to obtain component information from manufacturers it did not control. In our reading, that made the manufacturer’s willingness to disclose components a factor in whether a reseller could safely offer a product to the government.
Commercial and small purchases
The FAR Council applied the rule to contracts at or below the simplified acquisition threshold and to commercial items. The Administrator for Federal Procurement Policy applied it to commercially available off-the-shelf items. The rule said the risk was not reduced by an item being sold to the public or by a small purchase size. So a purchase card buy of a consumer security product was covered too.
What agencies still had to do
Contract clauses do not remove the agency’s duty. The statute prohibits use. If a contractor misses a component, the agency is still using it. Agencies needed their own checks alongside supplier reports, in our reading.
From report to decision
A contractor report is the start of an agency process, not the end. Someone has to read it, decide whether the article is covered, decide what to do with systems already using it and decide whether the contract needs a modification. The statute left no room for continued use after October 1, 2018. So the agency needed a decision path that could move from report to removal plan quickly, with the CISO, the contracting officer and counsel involved.
The ten-business-day follow-up report helped. It asked the contractor for mitigation details, the reasons the covered article was used and the steps taken to prevent recurrence. Those details are the raw material for the agency’s own decision and for any audit that followed.
How the ban landed on different public bodies
The statute applied to the Federal Government. Its effects reached further, through contracts, shared services and the example it set.
Large civilian departments
Large departments faced both instruments at once. They had to report branded products to DHS, start removal on the directive’s schedule and prepare for the statute’s wider scope. Their components ran separate IT estates, so one department might have dozens of inventories to reconcile.
Defense and the intelligence community
The directive did not reach national security systems or certain Defense and intelligence systems. The statute did. For those organizations, section 1634 was the first government-wide legal ban on the vendor’s products, in our reading. They also faced the statute’s embedded-component scope from the outset.
Small agencies, boards and commissions
For a small agency that relies on shared services, the use-through-others language mattered. Their exposure depended on the products used by their shared service providers. Their inventory work was partly a matter of asking those providers.
Legislative and judicial branch entities
The statute said no department, agency, organization or other element of the Federal Government. The directive applied only to executive branch agencies. In our reading, the statute’s wording was wide enough to reach legislative and judicial branch entities that the directive did not. Those bodies should check with their own counsel.
Agencies with large field operations
Agencies with offices across the country, or abroad, faced a logistics problem. Field offices may buy locally, run their own small servers and keep devices that seldom connect to headquarters networks. Discovery and removal had to reach each site. In our reading, those agencies needed site-level confirmation from field managers, backed by central tools where they could reach.
Shared service providers
Agencies that provide shared services to others carried a double load. They had to clean their own estates, and their customers depended on the result for compliance under the statute’s use-through-others language. A shared service provider that could document its component inventory gave every customer agency a ready answer. One that could not left each customer to ask separately.
State and local governments
On its terms, section 1634 applied to elements of the Federal Government, not to state, local, tribal or territorial governments. But they share networks, data and contractors with federal agencies. In our reading, a state agency running a federal program on federal systems would meet the ban through that system. A state agency buying for itself was outside the statute, though it could choose to follow it.
Inspectors general and auditors
Oversight bodies had their own role. An inspector general testing compliance would look for the agency’s method, its evidence and its record of findings, as described above. The statute’s subsection (c) also asked the Defense Department to report on capabilities and methods to monitor networks for prohibited products. In our reading, that request anticipated the question an auditor would ask: how does the agency know?
Grantees and research partners
Universities, research institutes and other grantees are not federal elements, so the statute did not bind them directly. But some run systems on behalf of agencies, and some deliver data or deliverables under contracts. Where a grantee or partner operates a system for an agency, that system can be a federal information system. Where it holds a contract, the acquisition clause applies. The answer depends on the instrument, which is why tracking by instrument matters.
Contractors and integrators
For federal contractors, the acquisition clause made the ban a contract term with reporting deadlines. Integrators that build systems from several vendors’ components faced the hardest discovery task. They had to ask each upstream supplier about its components and keep the answers.
Objections, in order
Five objections deserve an answer. We take them in the order they bear on the argument.
1. Few agencies ran Kaspersky products, so the scope question was academic
The volume of branded products in federal use is not something we can state from the sources. But the point of the in-whole-or-in-part phrase is that branded use is not the measure. An agency with no Kaspersky-branded product might still run a third-party product with an embedded engine. The scope question decided whether the agency had to look.
2. Suppliers would simply certify, so agencies did not need component inventories
Supplier certification was the main mechanism, and the 2018 clause relied on it. But the statute prohibited use by the government, so the risk of a wrong certification fell on the agency as user. An agency that could check certifications against its own records was better placed than one that could not.
3. The directive and the statute duplicated each other
They overlapped, but they were not the same. The directive reached fewer products and fewer systems, with earlier deadlines. The statute reached more products and more of government, later. The district court later found that the statute, not challenged in the directive suit, caused both injuries the company alleged from the directive. That finding is consistent with our reading that the statute was the wider rule.
4. Banning named vendors is a policy question, not an architecture one
Whether to ban a vendor is a policy question. How to comply is an architecture question. Section 1634 set the policy. Complying with it required knowing what was inside products, how services were built and what controls a product supported. Those are questions for enterprise architecture and asset management teams.
5. The ban was about one company, so its lessons do not carry over
Section 1634 named one vendor. But its subsection (c) asked for a review of government-wide authorities to remove suspect products, gaps in them and methods to monitor networks for prohibited products. Congress was asking how to do this again. The 2024 Commerce determination described below used the same in-whole-or-in-part language for a wider ban. The capability the statute assumed, knowing what is inside products, is not specific to one company.
What a leader could check
These checks follow from the four parts. They are our suggestions, not requirements.
- Can the agency list, for each security and network product, the licensed engines and major components inside it and who supplied them?
- Do contracts for managed and hosted services require the provider to disclose the tools it uses to deliver them?
- For each security product, is there a record of the controls it supports, so a removal plan can name the gaps it would open?
- Does the agency have a named owner and a procedure for acting on a contractor’s report of a banned component within days?
- Are vendor bans tracked by instrument, with the scope and dates of each recorded?
The checks need owners. Component inventories belong to the CIO’s asset management function. Supplier disclosures belong to the senior procurement executive’s staff. Control mapping belongs to the CISO. Acting on reports needs a joint process between the CISO, contracting officers and counsel.
The checks are worth running even where no vendor ban is in force. Each one tests a capability that any origin-based rule would demand: knowing components, knowing suppliers’ tools, knowing what controls depend on which product, and acting quickly on a report. Those capabilities also help an agency respond to a vulnerability in a widely licensed component, which raises the same question of where that component is.
With hindsight
The district court that later heard the vendor’s challenges filled in the December timeline. It counted the directive’s 90-day date as December 12, 2017, the day the statute was signed. It recorded that the Acting Secretary had issued a final decision on December 6, 2017, finding that the information gathered since issuance supported the original determination and maintaining the directive without modification.
Kaspersky Lab challenged both instruments in federal court in Washington. The first suit, filed on December 18, 2017, alleged that the directive and the December 6 final decision violated the Administrative Procedure Act and the company’s right to due process. A second suit alleged that the statute was an unconstitutional bill of attainder.
On May 30, 2018, the district court dismissed both, in Civil Action Nos. 17-2697 and 18-325. It held that the plaintiffs had not plausibly alleged that sections 1634(a) and (b) were a bill of attainder. It dismissed the directive suit for lack of standing. Even if the directive were rescinded, the court said, the plaintiffs’ alleged harms would continue in full force as a result of the statute. In our reading, that holding confirmed that the statute, not the directive, was the operative rule from then on.
The acquisition rule followed. On June 15, 2018, DoD, GSA and NASA published an interim FAR rule, effective July 16, 2018. It added FAR subpart 4.20 and the clause at 52.204-23, to be inserted in all solicitations and contracts. The clause defined a covered article to include products that contain components using hardware or software developed in whole or in part by a covered entity. It prohibited contractors from providing covered articles the government would use from October 1, 2018. It barred using them to develop data or deliverables first produced under the contract.
The clause’s reporting duty was specific. Contractors had to report a covered article within one business day of identifying it, with supplier, brand, model and mitigation details. They had to follow up within ten business days with further mitigation information and the efforts made to prevent use. The clause as it stands in the eCFR, dated December 2023, gives three business days for the first report and keeps the ten-business-day follow-up. The rule applied to small purchases, commercial items and commercially available off-the-shelf items. On September 10, 2019, the agencies adopted the interim rule as final without change.
A later confirmation of the component problem came from the Commerce Department. On June 24, 2024, its Bureau of Industry and Security published a final determination under its information and communications technology supply chain rule. It covered transactions in cybersecurity and antivirus products developed, manufactured or supplied in whole or in part by Kaspersky. It also covered the integration of Kaspersky software into third-party products, such as white-labeled products. The determination said white labeling increased the risks because the user would be less likely to know the true source of the code.
The determination set two dates. From July 20, 2024, Kaspersky could not enter new agreements with U.S. persons for the covered transactions. From September 29, 2024, it could not provide signature or codebase updates to U.S. users. Resale, integration into other products and licensing for those purposes were prohibited from the same date. Threat intelligence, security training and certain consulting services that are purely informational or educational were excluded.
CISA’s own page for the directive, as checked on November 15, 2026, says that Commerce added Kaspersky to the Entity List in 2024. It says the resulting prohibition applies more broadly than the directive, which contained exceptions.
What held up is the central point of the December 2017 statute: a ban defined by origin reaches inside other vendors’ products, and compliance depends on knowing components. The FAR clause spelled that out for contractors. The Commerce determination applied the same logic beyond government and named integration and white labeling explicitly.
What we cannot say from official sources is the number of agencies that found embedded components or how long removal took. Those results were reported to DHS and in agency submissions, not published in the sources we opened. The lesson we draw is narrower. A vendor ban is only as enforceable as the buyer’s knowledge of what its products contain.
Spotted an error? Tell us and we will check it and correct the article where needed.