Capital One’s breach surfaced through a stranger’s report
Capital One learned of its 2019 intrusion from an outside tip to its disclosure channel, months after the access. The path from an outsider’s report to a declared incident is a detection control and needs owners, look-backs and measures.
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
Capital One learned of its 2019 intrusion from an outside tip to its disclosure channel, months after the access. The path from an outsider’s report to a declared incident is a detection control and needs owners, look-backs and measures.
- The week On July 29, 2019, Capital One Financial Corporation announced a data security incident.
- The report as a detection signal A detection program can be described entirely in terms of tools: endpoint agents, network sensors, log analytics and the analysts who watch them.
- Intake that someone owns An intake path starts with a place to send reports and someone who reads them.
- Triage that looks backward Triage in a vulnerability program can stop at two questions: is the report valid, and how severe is it.
- The hand-off Once triage finds signs of use, the report becomes an incident.
- Evidence and measures A path that is never measured decays.
- With hindsight The bank’s regulator gave its answer to the cloud questions a year later.
The week
On July 29, 2019, Capital One Financial Corporation announced a data security incident. The company said that on July 19 it had determined there was unauthorized access by an outside individual. That person had obtained personal information about people who had applied for its credit card products and about its credit card customers. The company said it had immediately fixed the configuration vulnerability the individual exploited and had begun working with federal law enforcement. By the time of the announcement, the FBI had arrested the person responsible.
The scale was large. Based on its analysis to that point, Capital One said the event affected approximately 100 million people in the United States and approximately 6 million in Canada. The largest category was application data from consumers and small businesses who applied for its credit cards from 2005 through early 2019. That data included names, addresses, postal codes, phone numbers, email addresses, dates of birth and self-reported income. The individual also obtained customer status data such as credit scores, limits, balances and payment history, and fragments of transaction data from 23 days across 2016, 2017 and 2018.
The company said no credit card account numbers or log-in credentials were compromised. It put the exceptions at about 140,000 Social Security numbers of credit card customers and about 80,000 linked bank account numbers of secured card customers. For Canadian customers, approximately 1 million Social Insurance Numbers were compromised. Capital One said it encrypted its data as a standard, but that the unauthorized access also enabled decryption. Fields it tokenized, most notably Social Security and account numbers, used different methods and keys and remained protected.
The release also said how the bank found out. In the question-and-answer section of its release, Capital One said the configuration vulnerability was reported by an external security researcher through its Responsible Disclosure Program on July 17, 2019. It then began its own investigation, which led to the July 19 discovery. The access itself had occurred on March 22 and 23, 2019.
The U.S. Attorney’s Office for the Western District of Washington announced the arrest the same day, on a criminal complaint charging computer fraud and abuse. Citing the complaint, its release said the person charged had posted on GitHub about the theft, and that the intrusion occurred through a misconfigured web application firewall. On July 17, a GitHub user who saw the post alerted Capital One to the possibility of a data theft. After determining on July 19 that there had been an intrusion, Capital One contacted the FBI, and investigators identified the person who was posting about it.
Capital One said it expected the incident to generate incremental costs of approximately $100 million to $150 million in 2019, mainly for customer notifications, credit monitoring, technology and legal support. It said it carried insurance for certain costs of a cyber risk event, subject to a $10 million deductible. It said it would notify affected people through a variety of channels and make free credit monitoring and identity protection available to everyone affected. In answer to whether the vulnerability arose because it operated on the cloud, the company said this type of vulnerability was not specific to the cloud.
Much of the public debate about Capital One in 2019 turned on the cloud. That debate matters, and the bank’s regulator later passed judgment on how the bank had managed its move to the public cloud. This article is about something narrower and, we would argue, more transferable to other institutions: how the bank found out.
On the bank’s own account, the intrusion happened in March. The bank learned of it in July, because someone outside the company sent a report to its responsible disclosure channel. The Justice Department’s account adds that the report came from a person who had seen the intruder’s own public post. Neither public account says the bank’s internal controls raised the alarm. An outsider did, through a channel the bank had set up for a different purpose: reports of vulnerabilities, not of crimes already committed.
Our argument is that the path from an outside report to a declared incident is a detection control. It deserves the same engineering, ownership and evidence as an alert from a monitoring tool. The analysis falls into five parts:
- The report as a detection signal: why an outside report belongs in the detection program, and the strongest objection to treating it that way.
- Intake that someone owns: the channel, its scope, its legal terms and the people who read it.
- Triage that looks backward: turning “is this vulnerability real” into “has it already been used”.
- The hand-off: moving from a validated report to incident response, law enforcement and the notification clocks.
- Evidence and measures: what shows the path works, to a board, an examiner or an auditor.
The parts follow a report from arrival to proof. The first explains why the path matters. The next three trace it in order, and the last covers how an institution shows it works.
The report as a detection signal
A detection program can be described entirely in terms of tools: endpoint agents, network sensors, log analytics and the analysts who watch them. An outside report fits none of those categories, so it can be left out of detection metrics, budgets and testing. The Capital One case shows why it should not be.
What the tip did that the tools did not
The bank’s release says the configuration vulnerability was reported on July 17 and the intrusion discovered on July 19. The report did two things in sequence. It pointed to a weakness the bank had not found, and it led the bank to look for use of that weakness, which it then found.
The Justice Department’s account of the report is different in tone. It describes a GitHub user alerting the bank to the possibility of a data theft after seeing the intruder’s post. Either way, the bank’s detection came from outside, and through the channel it had built for vulnerability reports.
The point for other institutions is not that their tools are weak. Every detection method has blind spots, and an outside observer sees from a different angle. Researchers, customers, journalists, other companies’ security teams and law enforcement can all see evidence of an institution’s compromise before the institution does. A channel that reliably turns what they see into an investigation adds coverage that tools cannot buy.
Why tools can miss what an outsider sees
There is a general reason internal tools miss some intrusions, and it applies well beyond this case. Many detection methods look for activity that is unusual for a system or an account. When an intruder acts through access a system legitimately holds, such as a role, a service account or a misconfigured component trusted by the network, the requests can look like the system’s own work.
An outsider sees a different signal. A researcher sees the weakness from the internet. A person browsing a forum sees the data after it has left. Neither depends on the institution’s baselines. That difference in vantage point is what makes outside reports complementary to internal detection rather than a weaker copy of it.
Kinds of outside report
Outside reports arrive in several forms, and each needs a slightly different path:
- A vulnerability report from a researcher, about a weakness that may or may not have been used.
- A notice that the institution’s data has appeared somewhere it should not be, such as a public code repository, a paste site or a criminal forum.
- A report from a customer or partner of suspicious activity that suggests a compromise upstream.
- A notification from law enforcement or a government agency that it has seen indicators tied to the institution.
- A report from a vendor that its own compromise may have touched the institution’s data.
These are different events. What they share is that they come from outside, arrive unpredictably and go to whoever happens to receive them. Unless the institution designs a path for each, the path is whatever the recipient decides to do.
Where the expectation differs by sector
For federal agencies, the expectation is now explicit. The CISA directive applies to federal executive branch departments and agencies, with exceptions for national security systems and certain defense and intelligence systems. It does not bind state and local governments, banks or health systems.
Outside the federal government, a disclosure channel is a choice. A state agency, a county, a hospital or a credit union can adopt the same structure voluntarily, and the directive is a useful template because it is public and detailed. What an institution should not do is claim a regulation requires a disclosure program when its own rules do not say so.
The strongest objection
The strongest objection is that outside reports are a poor thing to rely on. Many reports to a disclosure channel may be low-value, duplicate or out of scope. The Capital One tip, on this view, was luck: an intruder who publicized the theft, and an observer who bothered to tell the bank. Building a program around luck seems backward. Investment should go to internal logging and monitoring that would have found the access in March.
The objection is right that internal detection comes first. It is wrong to treat the two as alternatives. We would expect a well-run intake path to cost little compared with detection tooling, and its value comes in exactly the cases where tooling failed. The CISA directive discussed in hindsight treats a disclosure policy as an essential element of vulnerability management. It also requires agencies to assess any past impact of a reported vulnerability. The directive reads as an official statement that outside reports and internal detection are parts of one system.
Intake that someone owns
An intake path starts with a place to send reports and someone who reads them. Both sound trivial. Both fail in ordinary ways.
The channel
Capital One had a Responsible Disclosure Program, which its release describes as an avenue for ethical security researchers to report vulnerabilities directly to the company. The report arrived there. An institution without such a channel would have received the same information, if at all, through a general inbox, a social media message or a call to customer service. Each of those routes adds delay and loss.
The federal directive that came later makes the point about channels explicit. It says that when reporters cannot find an authorized channel, they may turn to their own networks or look for security staff’s contact details online, or decide reporting is not worth the effort. A published security contact, monitored by people able to triage, is the minimum.
Scope and legal terms
A disclosure policy tells reporters what they may test and how. For a bank, scope decisions carry legal weight. Testing a production banking application can touch customer data, so the policy has to say what is allowed and what must be reported without further probing.
The legal terms matter as much as the scope. A reporter who fears prosecution may not report. The federal directive requires agency policies to commit not to recommend or pursue legal action against good-faith research that follows the policy. A bank cannot bind prosecutors, and its counsel should decide what it can promise. But the policy’s tone and terms decide whether the people most likely to see a problem first will tell the bank.
Reports that are not about vulnerabilities
The July 17 tip, on the Justice Department’s account, read less like a vulnerability report than a warning that data might already have been stolen. Channels designed for researchers can mishandle such reports. A triage team trained to ask for reproduction steps and a proof of concept may close a report that says only “your data is here”.
The intake procedure needs a separate category for reports of possible compromise, with a faster path. A report of a theft already under way is an incident lead, not a bug. It should reach the incident response team on the same day, without waiting for validation in the usual queue.
Systems nobody meant to expose
Scope has a trap. A disclosure policy lists the systems researchers may test, but the systems reported may be ones nobody listed. The federal directive addresses this by requiring agency policies to grow until they cover every internet-accessible system, including systems that were not intentionally made internet-accessible.
The directive phases this in. Beginning 270 days after issuance, agencies must add at least one system to scope every 90 days, until all such systems are covered within two years. For any institution, the lesson for intake is to handle reports about unlisted systems as seriously as reports about listed ones. A report about a forgotten test server may matter more than one about the main website.
Reports that arrive through other doors
Not every outside report reaches the security contact. Some arrive at customer service, the fraud team, investor relations, a branch or an executive’s inbox. A person who has seen stolen data may write to whoever they can find.
The intake procedure should therefore cover the whole institution, with the security mailbox as one door among several. Front-line staff need one instruction: a message that mentions stolen data, a security weakness or a hacking claim goes to the security contact at once, unedited. The fraud and customer service teams, which see many messages, benefit from a short script and a direct route.
Anonymity and the data inside reports
Reporters may not want to be identified. The federal directive requires agency policies to state that reporters may submit anonymously, and forbids requiring personally identifiable information. It also requires a statement prohibiting the disclosure to any third party of personally identifiable information that reporters discover.
For a bank or health system, the second point is the sharper one. A report of exposed data may contain customer or patient data as evidence. The intake channel then holds regulated data, and it needs the same handling controls as any other system that does. That means restricted access to the queue, retention rules and a way to delete evidence copies once the investigation no longer needs them.
Who reads it, and when
An intake channel is only as good as its coverage. A mailbox read during business hours by one person loses reports during vacations, weekends and holidays. We would set up a shared queue with named owners, coverage that matches the institution’s incident response hours, and an automatic acknowledgment that tells the reporter a person will respond.
The people who read reports also need authority. A triage analyst who suspects an active compromise should be able to open an incident without asking permission. A channel that routes everything through a product security team that has no incident authority adds a step at the worst time.
How the path differs by institution
A large bank may run a formal disclosure program with dedicated staff, a third-party platform and, sometimes, payment for findings. A community bank or credit union may have none of those. It still has an internet presence, and someone may still see its data in the wrong place.
For a smaller institution, the minimum is a published security contact, monitored by its security function or its managed security provider, and a written procedure for what happens next. Where a provider runs the monitoring, the contract should say who reads outside reports and how fast they reach the institution. The institution, not the provider, owns the decision to declare an incident.
Triage that looks backward
Triage in a vulnerability program can stop at two questions: is the report valid, and how severe is it. The Capital One sequence shows a third question that matters more: has the weakness already been used?
From validity to exploitation
When a vulnerability is reported, the natural instinct is to confirm it and fix it. Fixing it quickly is right. But a fix closes the door without telling the institution whether someone has already walked through it. The bank’s release describes both steps: it fixed the configuration vulnerability, and its investigation then led to the July 19 discovery of the intrusion.
We think the look-back is the step most at risk of being skipped. It needs different skills from the fix, because it means searching historical logs for signs of use rather than changing a configuration. And it produces no visible output if nothing is found. Under time pressure, a team that has fixed the weakness may consider the matter closed.
How far back to look
In this case, the access occurred in March and the report arrived in July. A look-back limited to the last few weeks would have missed it. The relevant window is the period during which the weakness existed, which may be far longer than the period during which anyone was watching.
That has a direct consequence for logging. An institution can only look back as far as its logs reach. If access logs for an internet-facing application are kept for thirty days, the look-back for that application is thirty days, whatever the exposure window. Log retention may be set by storage cost or by a general policy. It should also be set by how far back a look-back must reach for the systems that matter most.
What the logs must show
Retention is not enough; the logs must record the right things. For a misconfigured component like the one the Justice Department described, the useful evidence includes requests reaching the component, actions taken with any credentials it held, and access to the data stores behind it. Cloud platforms and on-premises systems both can produce such records, but in some cases only when they are switched on and routed to a place where they are kept.
Capital One’s release notes that the speed of its diagnosis was enabled by its cloud operating model. Whatever the platform, the general point holds. The look-back is fast when the logs exist, are retained and can be searched. It is slow or impossible when they cannot.
Fixing without destroying evidence
The fix and the look-back can pull against each other. Changing a misconfigured component quickly is right, but the change can overwrite the evidence of how the component was set up and what it allowed. Rotating credentials can make it harder to trace what those credentials did.
We would add a short preservation step before the fix: capture the configuration, export the relevant logs and snapshot the affected systems where that is feasible. The step takes minutes when it is scripted and planned. It takes much longer, or is skipped, when the team improvises under pressure.
Which protections held
A look-back has to establish more than whether data was reached. It has to establish which protections held, because that decides what was compromised and who must be told. Capital One’s release shows the point. It said the unauthorized access also enabled decryption of data the bank encrypted as a standard. Fields it tokenized, using different methods and keys, remained protected.
The same release said that, apart from about 140,000 Social Security numbers and about 80,000 linked bank account numbers, no Social Security or bank account numbers were compromised. So the look-back has to trace keys and tokens as well as files. An institution should know before an incident which data is tokenized, where the keys live and which identities can use them. Otherwise the look-back cannot answer the question that drives notice.
When the look-back finds nothing
Many look-backs may find no sign of use. That result still needs a record: the window searched, the logs used, the gaps in coverage and the confidence of the finding. A negative finding with known gaps is different from one with full coverage, and the difference matters if evidence of use appears later.
The record also protects the institution. If a weakness is later found to have been used, a documented look-back shows that the institution searched in good faith and where its visibility ended. Without one, the institution cannot show it looked at all.
Looking sideways as well as back
A reported weakness may not be unique. Capital One said that after discovering the vulnerability it verified there were no other instances in its environment and augmented its automated scanning to look for the issue continuously. Call it the sideways step: once a weakness is understood, search for the same pattern across the estate.
The Justice Department’s 2022 account, covered under hindsight, says the person convicted had used a scanning tool to look for misconfigured accounts and had downloaded data from more than 30 entities. A pattern exploited at one institution may be under active exploitation at others. So what is learned should be shared, through the channels the institution uses for threat intelligence, as well as to search internally.
The hand-off
Once triage finds signs of use, the report becomes an incident. The hand-off from the intake path to incident response is where we would expect outside reports to lose time, because two teams with different procedures meet.
Declaring the incident
The Capital One sequence moved fast: a report on July 17, a determination of intrusion on July 19, contact with the FBI, and an arrest and public announcement on July 29. Between them, the two releases set out that sequence.
For other institutions, the procedure should name the point at which triage hands to incident response and who decides it. A reasonable rule is that any credible sign of exploitation, or any report of data already exposed, opens an incident at once. The vulnerability ticket stays open alongside it, because the fix and the investigation are separate pieces of work.
Law enforcement
The Justice Department’s release credits the bank’s quick alert for allowing the FBI to trace the intrusion. An outside report that points to a person, an account or a public post is a lead that law enforcement can use, and it may be time-sensitive. A public post can be deleted and copied data can move.
The decision to contact law enforcement belongs in the incident procedure, with a named owner and a known contact. Institutions that build that relationship before an incident can make the call in hours. It also bears on notice timing wherever a notice law allows delay at law enforcement’s request.
The notification clocks
A declared incident starts notification clocks, and an outside report can start them sooner than an institution expects. The banking agencies’ 36-hour clock, covered under hindsight, runs from determination, not from the intrusion, and other notice laws may run from discovery. The faster the intake path turns a report into a determination, the sooner the clocks start, and the less time the institution has to understand what happened.
The answer is never to slow the intake path. The answer is to prepare the notification work so it can run alongside the investigation: templates, regulator contacts, counsel and a method for counting affected people by jurisdiction. Capital One’s July 29 announcement, which followed its July 19 determination, came with a count of affected people and categories of data, which shows how much of that work can be compressed when it is ready.
Two public accounts on one day
On July 29, the bank and the U.S. Attorney’s Office each issued a release. The two accounts agree on the key dates but describe the tip differently. The bank speaks of an external security researcher reporting a configuration vulnerability; the prosecutors speak of a person who saw a public post and warned of a possible theft.
The accounts do not contradict each other, but the difference shows a practical point. Once law enforcement is involved, an institution no longer controls the whole public story. Its own statement should be accurate on its own terms, consistent with what it has told law enforcement, and limited to what it knows. Counsel and communications should agree before the announcement which facts the institution will state and which it will leave to the investigators.
Customers in more than one country
The Capital One incident crossed a border. The release counted people in the United States and in Canada separately, listed Social Insurance Numbers as a Canadian data type and pointed to separate Canadian information pages in English and French. Each country brought its own notice rules, regulators and customer expectations.
An institution with customers in several jurisdictions needs its notice preparation to cover each of them. That includes the count of affected people per jurisdiction, the data types that matter in each, the language of the notices and the regulator contacts. The faster the intake path turns a report into a determination, the more this preparation is worth.
Finance in the incident
Outside reports that become large incidents quickly become financial events. Capital One’s release gave an expected range of incremental costs for 2019, said those costs would be reported as a separate adjusting item and described its insurance deductible. A public company has to be ready to say such things soon after it discloses an incident.
For the intake path, the implication is about who joins the incident and when. The finance function needs to be told early enough to estimate costs, check insurance terms and meet the insurer’s notice conditions. Those conditions are set in the policy, and an institution should know them before it needs them.
The reporter after the hand-off
Once an incident is open, the person who reported it can drop out of view. That is a mistake in two ways. The reporter may have more information, such as where they saw the data or what else they noticed. And a reporter who hears nothing may conclude the report was ignored and go public, which the federal directive names as a known risk.
The procedure should keep someone responsible for the reporter during the incident. That person should acknowledge, ask for more detail where useful, and say what can be said about progress. Counsel should review what is shared, especially where the report concerns stolen data.
Evidence and measures
A path that is never measured decays. The final part is how an institution shows, to itself and to those who supervise it, that outside reports become investigations quickly and reliably.
What to measure
The federal directive gives a ready list. Agencies must report, among other measures, the number of reports received and found valid, the median age of open valid reports, and the number older than 90 days. They must also report the median time to validate a report, to remediate it and to respond first to the reporter. Those measures were written for agencies, but they translate directly.
For a bank or health system, we would add two measures the directive does not name. The first is the time from receipt of a report suggesting compromise to the opening of an incident. The second is the share of validated vulnerability reports that received a documented look-back for past exploitation. These are our suggestions, not a requirement. They measure the two steps the Capital One sequence shows to be decisive.
Testing the path
Measures show how the path performs on real reports. Testing shows how it performs on the rare reports that matter. A simple exercise sends a realistic report to the public security contact outside business hours, saying that customer data has appeared on a public code repository. It then times each step: acknowledgment, routing to incident response, the decision to open an incident, the start of the look-back, and the decision about law enforcement.
Such an exercise is our suggestion, not a requirement. It tests coverage, authority and hand-off together, which no metric on routine reports can do. It also produces evidence an examiner or auditor can review.
What examiners and auditors can ask for
Bank examiners review information security programs against the interagency guidelines, the same guidelines the OCC cited in its 2020 action against Capital One. Those guidelines do not mention disclosure programs by name, and an institution should not claim they require one. But an examiner reviewing incident response and detection can reasonably ask how the institution receives and handles outside reports.
The useful evidence is a written intake procedure, the published contact, the triage categories, the look-back requirement, coverage records, the measures above and the results of exercises. For a health system or a public agency, the same file serves whichever incident rules apply to it, and the privacy and security officers should review it together.
Whether to pay for findings
Some institutions pay researchers for valid findings through a bug bounty. The federal directive distinguishes a disclosure policy from a bounty and does not require agencies to run one. It notes that a reward may attract people who would not otherwise look, and may also increase the number of low-quality submissions.
For a bank, health system or agency, we would put the disclosure channel and its handling path first. A bounty adds volume to that path, so it is worth adding only once the path can absorb more reports without slowing the ones that matter. The reports that matter most may come from people who expect no payment. The Capital One report arrived through a disclosure program.
What the path costs
We expect the path to cost less than many security controls, but it is not free. It needs coverage across the institution’s incident hours, triage staff who can tell a vulnerability report from a compromise lead, and log retention long enough for look-backs. It also needs time from incident responders, counsel and communications when a report turns out to matter.
The trade-off is mostly about attention. Every hour spent on low-value reports is an hour not spent on detection engineering. The answer is a triage design that sorts quickly: duplicates and out-of-scope reports closed with a standard reply, vulnerability reports into the normal queue, and compromise leads straight to incident response.
What a board should see
A board does not need report-level detail. It needs to know that outside reports reach the right people, that reports of compromise become incidents quickly and that look-backs happen. A short quarterly view can show the volume of reports and the share found valid, the median time to open an incident from a report of compromise, the share of valid reports with a documented look-back, and the result of the most recent exercise.
Those measures let a board ask whether the path is improving without needing to understand the technology behind it. They also give the board a factual basis for its oversight of incident response.
Vendors and the path
An institution may run internet-facing systems through vendors. A researcher who finds a weakness in a vendor-hosted application may report it to the vendor, to the institution or to neither. Contracts should say that the vendor will pass on reports affecting the institution, how fast, and that the vendor will support a look-back with its own logs.
The 2021 interagency rule covered in hindsight requires bank service providers to notify affected banking organization customers of certain incidents. Outside reports that never become incidents at the vendor may still matter to the bank. The contract is where that gap gets closed.
The Capital One breach is remembered for its size and its setting. For most institutions, the more useful memory is the sequence: an outsider saw it, a channel caught it, a team looked back and found it. Each of those steps can be designed, owned and measured, and none of them depends on where the systems run.
With hindsight
The bank’s regulator gave its answer to the cloud questions a year later. On August 6, 2020, the OCC assessed an $80 million civil money penalty against Capital One’s two national banks. It cited the bank’s failure to establish effective risk assessment processes before migrating significant technology operations to the public cloud, and its failure to correct the deficiencies in a timely way. The OCC said the deficiencies resulted in noncompliance with the Interagency Guidelines Establishing Information Security Standards. It also said it positively considered the bank’s customer notification and remediation efforts, and it issued a cease and desist order alongside the penalty.
Cloud providers added protections too. On November 19, 2019, AWS made version 2 of its EC2 Instance Metadata Service available. The announcement said version 1 would remain supported and that version 2 added protections against four types of vulnerabilities that could be used to reach the metadata service, naming open firewalls, reverse proxies and server-side request forgery. Under version 2, every request is protected by session authentication. The announcement did not tie the change to any particular incident, and we do not either.
The criminal case went to a jury. On June 17, 2022, the U.S. Attorney’s Office in Seattle announced that a jury had convicted the defendant of seven federal crimes: wire fraud, five counts of unauthorized access to a protected computer, and damaging a protected computer. The jury acquitted on access device fraud and aggravated identity theft. The release said the defendant had built a tool to scan Amazon Web Services accounts for misconfigurations and had downloaded data from more than 30 entities. It also stated that Capital One had settled customer lawsuits for $190 million.
The idea that outside reports belong in security management became official policy for federal agencies. On September 2, 2020, CISA issued Binding Operational Directive 20-01. It required each agency to set a monitored security contact for its .gov domains within 30 calendar days, and to publish a vulnerability disclosure policy within 180 calendar days. Within two years, every internet-accessible system had to be in scope. The directive requires handling procedures under which any current or past impact of a reported vulnerability is assessed and treated as an incident or breach, as applicable. It also sets quarterly metrics on report volumes and handling times. In May 2023, NIST published Special Publication 800-216, recommendations for federal vulnerability disclosure guidelines covering how reports are received, assessed and managed.
For banks, the nearest regulatory change we can point to was about notice rather than intake. On November 23, 2021, the OCC, the Federal Reserve Board and the FDIC published a final rule on computer-security incident notification. It requires a banking organization to notify its primary federal regulator of a notification incident as soon as possible and no later than 36 hours after determining one has occurred. It requires a bank service provider to notify each affected banking organization customer as soon as possible of an incident that has caused, or is reasonably likely to cause, a material service disruption or degradation for four or more hours. The rule took effect on April 1, 2022, with compliance required from May 1, 2022.
What held up, as we read it, is the narrow argument. Outside reports remain a channel through which institutions can learn of problems they missed, and the federal directive now treats the look-back step as mandatory for agencies. What changed is the cost of a slow path. With a 36-hour regulator clock running from determination, a bank whose intake path is improvised either determines late and explains why, or determines fast without the facts. Neither is a good position.
Spotted an error? Tell us and we will check it and correct the article where needed.