The 2015 DMCA security research exemption covered lawfully acquired devices, not systems one cannot acquire
The Librarian of Congress’s October 28, 2015 rule let researchers bypass access controls on lawfully acquired consumer devices, vehicles and implanted medical devices, with a 12-month delay for all but voting machines. A path for systems, with their owner’s or operator’s authorization, came in 2018.
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
The Librarian of Congress’s October 28, 2015 rule let researchers bypass access controls on lawfully acquired consumer devices, vehicles and implanted medical devices, with a 12-month delay for all but voting machines. A path for systems, with their owner’s or operator’s authorization, came in 2018.
- The week The rule described itself as the sixth triennial rulemaking under section 1201(a)(1) of title 17. [1] It said the Librarian’s determination rested on the Register’s Recommendation, transmitted on October 8, 2015. [1] It explained the statutory test.
- The scope: devices a researcher could acquire The exemption listed three kinds of device.
- The conditions: what a researcher had to satisfy The exemption applied to circumvention solely for good-faith security research. [1] The definition repeated “solely” for the access itself: accessing a computer program solely for good-faith testing, investigation or correction of a security flaw or vulnerability. [1] A mixed purpose, such as research combined with building a commercial product, sat uneasily with that word.
- The delay: twelve months, at other agencies’ request The rule recorded the positions of four agencies.
- The gaps: what the exemption did not address That left disclosure to other norms and laws.
- How the rule reached banks, health systems and agencies Health systems buy implanted and monitoring devices and use them in care.
- Objections, in order They can.
- Checks a CISO could run These checks are our suggestions, not requirements of the 2015 rule or of section 1201.
- With hindsight The next triennial rule, published on October 26, 2018, and effective October 28, 2018, revisited the security research exemption as Class 10. [2] Proponents asked to remove six features they said unnecessarily restricted the exemption or chilled research.
The week
On October 28, 2015, the Librarian of Congress published a final rule adopting exemptions to the Digital Millennium Copyright Act’s prohibition on circumventing technological measures that control access to copyrighted works. [1] The rule appeared at 80 FR 65944 and was effective the same day. [1] It was dated October 20, 2015, by both the Register of Copyrights and the Acting Librarian of Congress. [1]
The rule described itself as the sixth triennial rulemaking under section 1201(a)(1) of title 17. [1] It said the Librarian’s determination rested on the Register’s Recommendation, transmitted on October 8, 2015. [1] It explained the statutory test. Proponents must show that uses affected by the prohibition are or are likely to be noninfringing. They must also show that the prohibition is causing, or in the next three years is likely to cause, an adverse impact on those uses. [1]
The Copyright Office had grouped the petitions it accepted into twenty-seven proposed classes of works. [1] Three of them concerned security research. Proposed Class 25 sought an exemption for good-faith research into flaws in computer programs on all types of systems and devices. [1] Proposed Class 22 covered software in motorized land vehicles, for researching their security or safety. [1] Proposed Class 27A covered software in implanted medical devices and their monitoring systems. [1]
The rule recorded the arguments. Proponents said software is pervasive in machines and devices, including vehicles, home appliances and medical devices, and that independent research is needed to find flaws. [1] They said the permanent statutory exemptions for reverse engineering, encryption research and security testing, in sections 1201(f), (g) and (j), did not give enough assurance. [1] Opponents, a range of companies and organizations representing copyright owners, argued the uses were not shown to be noninfringing, that the statutory exemptions sufficed and that any exemption should include a disclosure requirement. [1]
The Copyright Office notified the Department of Transportation, the Environmental Protection Agency and the Food and Drug Administration. [1] The rule said all three responded with significant reservations about potential effects on public health and safety. [1] The National Telecommunications and Information Administration fully supported a broad exemption for all computer programs, regardless of device. [1]
The Register recommended a narrower exemption than Class 25 sought. The rule said the record focused on consumer-facing products rather than large-scale industrial or government systems, and included evidence on motor vehicles, implanted medical devices and electronic voting machines. [1] The Register concluded the record did not support an open-ended exemption covering highly sensitive systems such as nuclear power plants and air traffic control. [1]
The Librarian adopted the Register’s version. It allowed circumvention of access controls on computer programs on a lawfully acquired device or machine, solely for good-faith security research. The circumvention could not violate any applicable law, including the Computer Fraud and Abuse Act. [1] The device had to fall in one of three categories. It could be primarily designed for individual consumers, including voting machines, or be a motorized land vehicle. Or it could be an implanted medical device or its personal monitoring system that is not and will not be used by patients or for patient care. [1]
The exemption carried a condition on timing for every device except voting machines. Circumvention could be “initiated no earlier than 12 months after the effective date of this regulation.” [1] The rule said the Register found twelve months was the shortest period that would reasonably let other agencies respond, and recommended immediate effect for voting machines because no public safety issue justified delay there. [1]
The rule defined “good-faith security research” as accessing a program solely for good-faith testing, investigation or correction of a security flaw or vulnerability. [1] The activity had to be carried out in a controlled environment designed to avoid harm to individuals or the public. [1] The information derived had to be used primarily to promote the security or safety of the class of devices or their users, and not used or kept in a way that facilitates copyright infringement. [1]
The 2015 security research exemption is remembered as the moment researchers won legal room to examine cars, pacemakers and voting machines. [1] For technology and security leaders at banks, health systems and public agencies, its boundary is more instructive than its breadth. The exemption covered devices a researcher could lawfully acquire. [1] In our reading, it did not cover the systems those institutions run.
Our argument is that the 2015 exemption drew its line at lawful acquisition of a device, and that this line left institutional systems outside it. The test was a lawfully acquired device of a listed kind. [1] A bank’s online banking platform, a hospital’s network of infusion pumps or an agency’s benefits portal is not a device a researcher acquires. In our reading, the 2015 rule gave consumer-device research a legal path and left research on institutional systems where the statute already had it.
That matters because institutions are on both sides of security research. They buy devices researchers test, and they operate systems researchers may probe. They also commission research on their own systems. Each role meets the exemption differently.
We take the 2015 exemption in four parts:
- The scope: which devices and programs it covered.
- The conditions: what a researcher had to satisfy to rely on it.
- The delay: why much of it waited twelve months, and who asked for that.
- The gaps: what the exemption did not address at all.
The parts do not overlap. The first is coverage, the second eligibility, the third timing and the fourth omissions. After them we look at how the rule reached banks, health systems and agencies, take the objections in order and end with checks a CISO could run.
The scope: devices a researcher could acquire
Three categories
The exemption listed three kinds of device. The first was a device or machine primarily designed for use by individual consumers, with voting machines expressly included. [1] The second was a motorized land vehicle. [1] The third was a medical device designed for whole or partial implantation in patients, or a corresponding personal monitoring system, that is not and will not be used by patients or for patient care. [1]
Each category is defined by what the device is, not by who is studying it. A researcher at a university, a company or acting alone could rely on the exemption if the device and the conduct fit. The 2015 text placed no condition on the researcher’s identity. [1]
Why the record decided the line
The rule explained the narrowing as a matter of evidence. The Class 25 proposal was very broad and potentially swallowed the vehicle and medical device classes, the rule said, but the record focused primarily on consumer-facing products. [1] The rule set that focus against large-scale industrial or government systems such as power or transit systems. It said the record did not support an open-ended exemption reaching highly sensitive systems such as nuclear power plants and air traffic control systems. [1]
That reasoning matters for institutions. The exemption’s scope followed the petitioners’ evidence. Researchers had shown harm from access controls on consumer devices, cars, implanted devices and voting machines. [1] They had not, on this record, shown the same for enterprise or infrastructure systems. In our reading, the omission of institutional systems was a finding about the record, not a judgment that research on them was unwelcome.
Lawful acquisition as the anchor
Circumvention had to be undertaken on a lawfully acquired device or machine. [1] Acquisition implies possession of a copy. A researcher could buy a consumer router or a car, or obtain an implantable device not intended for patient use. A researcher cannot acquire a bank’s core system or a hospital’s electronic health record.
The anchor also excluded devices in service. The medical device category required that the device not be, and not be going to be, used by patients or for patient care. [1] The rule said this followed FDA’s suggestion. [1] A pump on a hospital ward was outside the exemption by its terms. Only a device set aside from care could be studied under it.
Programs, not data
The exemption’s object was computer programs. It allowed circumvention of access controls on computer programs operating on the covered devices. [1] Its definition spoke of accessing a computer program for testing, investigation or correction of a flaw. [1] The use condition barred using or keeping the results in a way that facilitates copyright infringement. [1]
For an institution, that framing is a reminder of what the exemption did not touch. It addressed the copyright barrier to studying software. It did not address privacy law, data protection duties or contractual limits on what a researcher could do with personal data found on a device. A hospital that sent a used monitoring device for research, for instance, still had to deal with any patient data on it under its own obligations. The exemption was silent on that, as a copyright rule would be.
Vehicles, two ways
Vehicles appeared in two separate proposals. Proposed Class 22 sought circumvention of access controls on vehicle software for researching security or safety, undertaken by or on behalf of the vehicle’s lawful owner. [1] Proposed Class 21 concerned vehicle software for diagnosis, repair or modification, and the exemption adopted under it named personal automobiles, commercial motor vehicles and mechanized agricultural vehicles. [1] The rule handled security research through the combined security exemption, with motorized land vehicles as one of its three device categories. [1]
For public agencies and companies that run fleets, the vehicle category is the one likely to touch their own property. A fleet vehicle is lawfully acquired and is a motorized land vehicle. In our reading, the 2015 exemption reached research on such vehicles, under its conditions, in a way it did not reach the fleet operator’s dispatch or telematics servers.
Voting machines as a consumer device
The rule placed voting machines inside the consumer-device category by name. [1] It also exempted them from the twelve-month delay. [1] For election officials, that put independent research on voting equipment expressly within an exemption from the access-control prohibition from the rule’s effective date, subject to the same conditions as any other device.
The conditions: what a researcher had to satisfy
“Solely” for good-faith research
The exemption applied to circumvention solely for good-faith security research. [1] The definition repeated “solely” for the access itself: accessing a computer program solely for good-faith testing, investigation or correction of a security flaw or vulnerability. [1] A mixed purpose, such as research combined with building a commercial product, sat uneasily with that word.
The 2018 rulemaking record later described this as the access limitation, one of six features proponents asked to remove. [2] It was not removed then, as the hindsight section shows. [2]
What “good faith” did and did not settle
The rule used “good-faith” as a defined term rather than as a test of motive. It defined good-faith security research by purpose, setting and use of results. [1] It did not ask whether the researcher intended to notify the manufacturer, how long the researcher would wait, or whether the researcher would accept payment.
A definition built that way is easier to apply and harder to dispute. A researcher could show the purpose, the environment and how results were used. Motive is harder to prove. In our reading, the rule chose observable conditions over intentions, which also tells an institution what evidence a researcher relying on the exemption would be expected to keep.
A controlled environment
Research had to be carried out in a controlled environment designed to avoid any harm to individuals or the public. [1] The rule said this answered other agencies’ concern that research must not put the public at risk. [1] In our reading, for vehicle research that pointed to a test track rather than a public road, and for medical devices to a bench, not a patient.
For an institution, the condition is familiar. It is the same principle that governs penetration testing in production: define the environment, the limits and the safeguards before testing begins. The 2015 rule wrote that principle into the copyright exemption.
Use of the results
The information derived had to be used primarily to promote the security or safety of the class of devices, or of those who use them. It could not be used or kept in a way that facilitates copyright infringement. [1] The rule said the Register drew this from what she saw as the basic intent of section 1201(j), the statutory security testing exemption. [1]
“Primarily” left room for other uses, such as publication or teaching. It did not say how findings should be shared. That question, as discussed below, the rule deliberately left open.
Who bears the risk of a misjudged condition
The exemption’s conditions are judgments. Whether an environment was designed to avoid harm, or results were used primarily for security, can be argued after the fact. [1] The researcher relying on the exemption bore the risk of getting them wrong. A manufacturer that disagreed could contend the conduct fell outside.
That allocation shaped behavior on both sides. Researchers had reason to document their setting and their handling of results. Manufacturers had reason to state publicly how they would treat good-faith reports. Institutions that bought devices had an interest in both: they benefited when researchers found flaws and when manufacturers fixed them without a dispute about whether the research was permitted.
Leased and subscribed equipment
Lawful acquisition sounds simple. Institutions complicate it. Hospitals and agencies lease equipment, buy devices under service agreements and license the software that runs on them. Whether research on such equipment is on a “lawfully acquired device or machine” is a question the 2015 text did not address beyond the words themselves. [1]
The 2018 record shows the question stayed live. Proponents challenged the lawfully acquired limitation, and the Justice Department’s computer crime section said it considered the existing language preferable to conditioning the exemption on ownership of a particular copy of software. [2] The limitation was kept. [2] For an institution that wants research done on leased equipment, the lease terms and the vendor’s position matter as much as the exemption.
The conditions as a test plan
Read together, the conditions resemble the rules of engagement an institution writes for a penetration test. There is a scope, here a lawfully acquired device of a listed kind. [1] There is a purpose, testing, investigation or correction of a flaw. [1] There is an environment designed to avoid harm. [1] There is a rule for results, used primarily for security or safety. [1] And there is an express reservation of other law. [1]
An institution that commissions device research can use the same structure. A statement of work that names the devices, the purpose, the test setting, the handling of findings and the legal review maps onto each condition. That does not make the institution’s contract a copyright safe harbor. It does make the record of the engagement easier to compare against the exemption if questions arise.
Other laws, named
The exemption applied only where circumvention does not violate any applicable law, including without limitation the Computer Fraud and Abuse Act of 1986. [1] The rule said the Register tracked Congress’s language in section 1201(j) rather than proponents’ alternatives, including by expressly excluding acts that violate other laws such as the CFAA. [1]
In our reading, this condition matters a great deal for institutional systems. A researcher testing a system the researcher does not own or control faces the CFAA regardless of copyright. The 2015 exemption did not change that. It made compliance with other laws a condition of the copyright exemption itself, so that a CFAA problem could also become a section 1201 problem.
The delay: twelve months, at other agencies’ request
What the agencies said
The rule recorded the positions of four agencies. DOT, EPA and FDA expressed significant reservations about public health and safety. [1] DOT, for example, was concerned that independent researchers may not fully appreciate the consequences of circumvention for vehicle safety or the logistics of remedial action. [1] NTIA supported a broad exemption and believed other agencies’ concerns could be addressed by stating that the exemption does not obviate compliance with other laws. [1]
The Register noted that different parts of the Administration appeared to hold divergent views on security research and that the exemption could cover any number of highly regulated products. [1] She recommended a delay to give other parts of the government an opportunity to respond. [1]
What the delay was for
The rule framed the twelve months as time for other agencies to consider and prepare. [1] NTIA had acknowledged that a delay could be appropriate for that purpose. [1] The rule did not say what the agencies would do with the time.
For institutions, the delay meant the exemption’s practical start for vehicles, medical devices and consumer devices other than voting machines was the date the rule set by formula, not the publication date. Consider a medical device maker or a hospital security team reading the rule in October 2015. Each knew that research under the exemption on those devices could not be initiated earlier than “12 months after the effective date of this regulation.”
How institutions could use the year
For a health system or an agency, the delay was time to prepare for research that would begin on devices they used. Several steps were within reach. An inventory of implanted-device programs and the monitoring systems patients took home would show which products fell in the exemption’s medical category. A list of vehicles and consumer devices in service would do the same for the other categories.
A second step was intake. Once research could proceed, findings would arrive through manufacturers, regulators or publication. An institution with a named owner for device security advisories would be ready to assess them. One without would learn of a flaw in its devices from whoever read the news first.
What the delay signaled
The delay was a signal about where regulators stood. Safety agencies wanted time. The copyright office and the telecommunications agency were prepared to proceed. In our reading, the rule’s text records an unresolved policy disagreement inside the government about independent security research on safety-critical products, resolved for the moment by a date.
The gaps: what the exemption did not address
Disclosure
Opponents asked for an express disclosure requirement, so that a developer or manufacturer would have time to fix a flaw before it became widely known. [1] The rule said the Register supported responsible disclosure but acknowledged the difficulty of defining disclosure standards in the rulemaking, as opinions were sharply divided. [1] She did not include a disclosure rule. [1]
That left disclosure to other norms and laws. For a manufacturer, it meant the exemption gave no guaranteed window before publication. For a health system or bank that bought the device, it meant learning of a flaw could come from a researcher, the manufacturer or a public report, in any order.
Tools and services
The rule noted that exemptions under section 1201(a)(1) apply only to the act of circumventing an access control. [1] Other parts of section 1201 address the manufacture and provision of, or trafficking in, products and services. [1] The exemption did not change those provisions.
For institutions that commission research, this is a practical point. An exemption for the researcher’s own act of circumvention does not, by itself, authorize a vendor to supply circumvention tools. In our reading, organizations hiring testers to examine devices needed to consider the trafficking provisions separately, with counsel.
No role for the manufacturer’s consent
The 2015 exemption required lawful acquisition of the device. [1] It did not require the manufacturer’s consent, notice to the manufacturer before research began, or a disclosure timeline. [1] Opponents had argued an exemption was unnecessary partly because manufacturers frequently authorize independent research. [1] The rule did not make that authorization a condition.
That is the feature that made the exemption meaningful for consumer devices. Research could proceed on a lawfully acquired device without the maker’s approval. For institutional systems, the opposite logic applied, as the 2018 rule later made explicit: research there would require the owner’s or operator’s authorization. [2] In our reading, the two approaches fit their objects. A device sold to the public can be studied by anyone who buys it. A system run by an institution is the institution’s to open.
The statutory exemptions left in place
The 2015 rule did not displace the permanent exemptions in section 1201. It named three: reverse engineering in 1201(f), encryption research in 1201(g) and security testing in 1201(j). [1] The Register concluded those provisions were inadequate for the proposed research because of limitations and conditions in them. [1]
That conclusion matters for institutions because the statutory exemptions are where research on their systems would have to fit, absent coverage in the rule. In our reading, the Register’s finding that 1201(j) was inadequate for consumer-device research gave institutions no reason to assume it was adequate for research on theirs. The rule left that question unaddressed.
Fair use as the foundation
The Register concluded that good-faith security research into programs operating the covered devices is likely a noninfringing fair use, and for vehicle software may be noninfringing under section 117. [1] She also found that access controls had hindered legitimate research on those programs. [1] Those findings were the basis for the exemption. [1]
The fair use finding was tied to the record. It covered the programs on the devices before the Office. A later proceeding with evidence about other systems could reach a different scope, as the hindsight section shows happened in 2018. [2]
Institutional systems
The largest gap, for this audience, was coverage of systems institutions operate. The exemption applied to lawfully acquired devices of three kinds. [1] An enterprise network, a payment platform or a records system run by an agency was outside it. In our reading, research on those systems still depended on the statutory exemptions and on the operator’s permission, as it had before. [1]
The permanent statutory exemption for security testing in section 1201(j) remained. [1] Proponents argued it gave insufficient assurance for their purposes, and the Register agreed for the devices in the record. [1] For institutional systems, the record did not reach the question.
Patient data
The medical device petition also sought a separate exemption for patients to access data generated by their own implanted devices. [1] The Office treated that as a separate class, Proposed Class 27B, distinct from security research. [1] We note it to mark the boundary: the security research exemption itself did not address patients’ access to their own device data.
How the rule reached banks, health systems and agencies
Health systems
Health systems buy implanted and monitoring devices and use them in care. The exemption covered research on implanted devices and their personal monitoring systems not in use for patient care. [1] It did not cover devices in clinical use. [1] In our reading, a health system’s own security team, testing devices on its network, could not rely on the 2015 exemption for devices in service.
The exemption still changed the information a health system could expect. Independent researchers could, after the delay, examine devices of the kinds the hospital used, in controlled settings. [1] Findings about those devices would reach the hospital through manufacturers, regulators or publication. A hospital that tracked device security needed a way to receive and act on that information.
Banks and credit unions
A bank’s systems are institutional. Its core banking platform, online banking and payment systems are not consumer devices a researcher acquires. [1] In our reading, the 2015 exemption’s categories did not cover them, and research on them continued to depend on the bank’s authorization and on laws such as the CFAA. [1]
Banks also issue consumer-facing devices and software, such as mobile apps running on customers’ phones. A researcher testing a banking app on a phone the researcher owns is working on a consumer device. In our reading, whether the 2015 exemption reached that research turned on how the app’s access controls related to the phone, a question for counsel rather than for this article.
Public agencies
Election officials saw a direct change. Voting machines were named as consumer devices and exempt from the delay. [1] Independent researchers could examine voting equipment they lawfully acquired, in a controlled environment, from the rule’s effective date. [1]
Other agencies operate systems the rule placed outside its scope. The Register expressly declined to cover highly sensitive systems such as nuclear power plants and air traffic control. [1] Agency networks and benefits systems fell outside for the same reason as bank systems: they are not devices a researcher acquires.
Vendors that host institutions’ systems
Banks, health systems and agencies may run important systems on vendors’ platforms. Those platforms are not devices a researcher acquires either. [1] In our reading, under the 2015 rule, research on a hosted platform depended on the vendor’s permission, not the institution’s, and on the statutory exemptions and other laws.
That creates a gap in authority. An institution may want its hosted application tested. The vendor controls the platform. In our reading, contracts that give the institution a right to commission testing, and set how the vendor will authorize it, were the practical answer, then and now.
Security teams as buyers
Every institution is a buyer of devices and software. The 2015 exemption changed the buyer’s information position for the covered categories. After the delay, independent research on those devices had an express exemption behind it. [1] Findings that a manufacturer might not volunteer could reach the market through researchers.
Buyers can use that. Procurement questions about how a manufacturer receives and handles outside research, and what it has fixed, become more answerable when outside research is happening. In our reading, the exemption gave institutional buyers a reason to ask device makers about their vulnerability handling as a condition of purchase.
Smaller institutions
A community bank, a rural hospital or a small county does not run a research program. For such an institution, the exemption’s effect was indirect. It may not commission device research or hear from researchers about its own systems. Its exposure was to findings about the products it bought.
That makes the vendor relationship the small institution’s main tool. A clause requiring the vendor to notify customers of security findings in its products, whatever their source, does more for a small institution than any internal research capability could. In our reading, the exemption raised the value of such clauses for buyers of every size.
Universities and academic medical centers
Academic institutions sit on both sides at once. They employ researchers who study devices, and some run hospitals that use them. The 2015 record included submissions from academic researchers seeking the broad Class 25 exemption. [1] An academic medical center could find its own faculty publishing research on devices its clinicians use.
For such an institution, internal coordination is the issue. A research group working under the exemption, in a lab, on devices not used for patient care, is within the rule’s conditions. [1] The hospital’s security and clinical engineering teams need to hear its findings before or when they become public. In our reading, the exemption gave academic institutions a reason to connect their research offices and their hospital security functions.
Device makers
Vehicle makers and medical device manufacturers were directly affected. Several of their trade associations opposed the classes. [1] The exemption meant independent researchers could study their products after the delay, subject to the conditions. [1] For manufacturers, a channel for receiving researchers’ findings became more useful once research could proceed with a copyright exemption behind it.
Objections, in order
“Institutions can authorize research on their own systems anyway”
They can. An institution that hires a tester or runs a vulnerability program authorizes access to its systems. The 2015 exemption did not need to cover that, and the hindsight section shows the 2018 rule added an express path for research with the owner’s or operator’s authorization. [2] The argument does not depend on institutions lacking permission. It depends on what the 2015 rule did not do for research the institution had not commissioned.
“The exemption was about copyright, not security”
Formally, yes. Section 1201 is a copyright provision. [1] The record, though, was about security. Proponents argued security flaws pose threats of physical injury, property damage and financial harm. [1] Safety agencies weighed in. [1] The rule’s conditions, a controlled environment and use of results primarily to promote security or safety, are security conditions. [1] In our reading, the copyright form carried a security policy.
“A twelve-month delay was harmless”
The delay was the Register’s compromise with agencies that had concerns. [1] It may have been reasonable. It also meant research the Register found likely noninfringing waited twelve months for every covered device except voting machines. [1] Whether that was harmless depends on what flaws existed in the meantime, which the record could not show.
“A narrow scope protected safety”
This is the strongest case for the 2015 line. Safety agencies had reservations, and the Register limited the exemption to the record and excluded devices in patient care. [1] Leaving out power plants and air traffic control kept research away from systems where a mistake could cause wide harm. [1]
The argument here accepts that. Its point is narrower: the line that protected those systems also excluded every institutional system, including ones where testing with the operator’s consent carries ordinary risk. The 2018 rule later drew a different line, using authorization rather than acquisition, which suggests in our reading that safety did not require the device test alone. [2]
“Conditions on paper do not change conduct”
A skeptic could say researchers who ignore the law would ignore the exemption’s conditions too. That is likely true and beside the point. The exemption was for researchers who wanted to stay within the law. [1] For them, the conditions defined how to work: on acquired devices, in safe settings, with results used for security. Institutions dealing with researchers could expect those who relied on the exemption to work that way, and could ask them to show it.
“Without a disclosure rule, the exemption invited reckless publication”
Opponents argued this. [1] The rule’s answer was the use condition: results had to be used primarily to promote security or safety. [1] That is a purpose test, not a timeline. For institutions that depend on devices, the absence of a disclosure rule placed weight on their own monitoring and on manufacturers’ processes for receiving reports.
Checks a CISO could run
These checks are our suggestions, not requirements of the 2015 rule or of section 1201. Organizations should confirm legal questions about research and circumvention with counsel.
- Authorization on record: does every test of the organization’s systems, internal or external, rest on a written authorization stating scope, environment and limits?
- A front door for findings: is there a published way for outside researchers to report vulnerabilities, with an owner who acknowledges reports?
- Device intelligence: for medical devices, vehicles and consumer devices the organization uses, who tracks published research and manufacturer advisories?
- Test environments: when the organization commissions research on devices, are they tested outside service, in an environment designed to avoid harm?
- Tool sourcing: when testers use circumvention tools, has counsel considered the separate trafficking provisions of section 1201?
- Hosted platforms: do contracts with hosting and software vendors give the organization a right to commission security testing, and say how the vendor authorizes it?
- Device inventory: does the organization know which of its devices fall in categories researchers can study, such as implanted-device programs, fleet vehicles and consumer devices?
- Election equipment: for agencies that run elections, is there a process to receive and assess independent research on the voting equipment in use?
The 2015 rule put device research on a legal footing and left institutional systems to the operator’s permission. [1] The checks above follow that line: make permission explicit where the institution controls the system, and make reception reliable where it does not.
One more question ties the checks together. If an outside researcher contacted the organization tomorrow about a flaw in a system it runs, who would answer, and what record would show what the organization had and had not authorized? The answer is the organization’s real policy on security research, whatever its written policy says. The 2015 rule did not ask institutions that question. Its successors, by adding a path that turns on authorization, in our reading effectively do. [5]
With hindsight
The next triennial rule, published on October 26, 2018, and effective October 28, 2018, revisited the security research exemption as Class 10. [2] Proponents asked to remove six features they said unnecessarily restricted the exemption or chilled research. They were the device, lawfully acquired, access, controlled environment, use and other laws limitations. [2]
The rule recorded a letter from the Justice Department’s Computer Crime and Intellectual Property Section supporting several changes, including eliminating the language confining the exemption to devices primarily designed for consumers. [2] It said the section recommended keeping the express reference to the Computer Fraud and Abuse Act. [2] NTIA again recommended an exemption regardless of the device on which programs run. [2]
The Acting Register recommended removing the device limitation. [2] She also recommended allowing circumvention on a “computer, computer system, or computer network on which the computer program operates,” undertaken with the authorization of its owner or operator. [2] The rule said this addressed research on structures, such as a building automation system, that cannot be acquired in the sense of physical possession. [2] It dropped “controlled” from the environment condition, and kept the other limitations. [2]
The 2021 rule, published and effective on October 28, 2021, considered further expansion as Class 13. [3] It said the Register recommended removing the condition that circumvention not violate other laws. In its place would be a statement that the exemption is not a safe harbor from liability under them, a change the Justice Department supported. [3] The Register did not recommend removing the access and use limitations or the lawfully acquired requirement, and found it unnecessary to name privacy research expressly. [3]
The 2024 rule, published and effective on October 28, 2024, renewed the exemption. [4] It said no oppositions were filed against renewal, and that petitioners highlighted research on voting machines, devices underpinning the financial industry and smart phones. [4]
As checked on November 16, 2026, the eCFR text of 37 CFR 201.40, up to date as of November 12, 2026, sets out the exemption at paragraph (b)(18). [5] It allows circumvention on a lawfully acquired device or machine, or on a computer, computer system or network with the authorization of its owner or operator, solely for good-faith security research. [5] It keeps the definition’s requirements of an environment designed to avoid harm and use of information primarily to promote security or safety. [5] It states that qualifying research may still incur liability under other laws, including the Computer Fraud and Abuse Act, and that eligibility is not a safe harbor from, or defense to, that liability. [5]
What held up from October 2015 is the shape of good-faith research. The definition’s core, access solely for testing, investigation or correction, an environment designed to avoid harm and use of results primarily for security or safety, is in the current text. [1] [5] What did not hold up is the device list. The device categories were removed in 2018, and research on systems a researcher cannot acquire now has a path through the owner’s or operator’s authorization. [2] [5]
In our view, the change made institutions’ authorization records part of the legal structure of research on their systems. An exemption that turns on the owner’s authorization works only where the owner can show, and the researcher can rely on, what was authorized. [5] Since 2018, the line at lawful acquisition of a device has stood beside a second line at permission.
Spotted an error? Tell us and we will check it and correct the article where needed.