Intel’s 2017 AMT flaw lived beneath the operating system, and its fix waited on each PC maker’s firmware

On May 1, 2017, Intel disclosed a critical flaw in the management firmware of business PCs. In our reading, it showed that patch programs built around operating systems could not see, test or deliver a fix that had to come from each computer maker.

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

On May 1, 2017, Intel disclosed a critical flaw in the management firmware of business PCs. In our reading, it showed that patch programs built around operating systems could not see, test or deliver a fix that had to come from each computer maker.

  1. The week These are remote management features that Intel supplies for business computers.
  2. The layer: a management engine below the operating system The first part asks where the flaw lived.
  3. The inventory: four facts about each machine The second part asks what an organization needed to know before it could act.
  4. The supply chain: one fix, many delivery schedules The third part asks how the fix reached the machine.
  5. The interim: living with exposure until a fix arrived The fourth part asks what an organization could do before its maker shipped a fix.
  6. How the flaw reached different organizations The four parts applied everywhere, and they played out differently by sector.
  7. Objections, in order Intel limited the network path to provisioned AMT and ISM systems. [1] That narrowed the most severe path.
  8. What infrastructure and security leaders could check The four parts translate into checks any organization can run on the firmware layer today, for this flaw or the next one in the same class.
  9. With hindsight The AMT flaw turned out to be the start of a sequence.

The week

On May 1, 2017, Intel published security advisory INTEL-SA-00075. [1] It rated the issue Critical and described its impact as elevation of privilege. [1] The advisory covered three product families: Intel Active Management Technology (AMT), Intel Standard Manageability (ISM) and Intel Small Business Technology (SBT). [1]

These are remote management features that Intel supplies for business computers. CERT/CC described them as hardware-based technologies meant for maintenance of computer systems, supported by the Intel Management Engine. [2] They listen for remote commands on known network ports. CERT/CC listed ports 16992 and 16993 for the web interface, and noted that 16994, 16995, 623 and 664 may also be used. [2]

Intel’s summary was short. The flaw could let an unprivileged attacker gain control of the manageability features these products provide. [1] It affected manageability firmware versions 6.x, 7.x, 8.x, 9.x, 10.x, 11.0, 11.5 and 11.6. [1] Versions before 6 or after 11.6 were not affected. [1] Intel’s resolved-firmware table mapped those versions to processor generations, from 1st Gen Core to 7th Gen Core. [1]

The advisory described two ways in. In the first, an unprivileged network attacker could gain system privileges on provisioned AMT and ISM systems. Intel scored that path at CVSS 9.8, Critical. [1] In the second, an unprivileged local attacker could provision manageability features and gain network or local system privileges on AMT, ISM and SBT. Intel scored that path at 8.4, High. [1] SBT was not exposed to the first path. [1]

Intel also said what was out of scope. The flaw did not exist on Intel-based consumer PCs with consumer firmware. [1] It did not exist on Intel servers using Intel Server Platform Services, or on certain Xeon workstations using that firmware. [1]

The recommended response had four steps. First, determine whether a system was AMT, SBT or ISM capable. Second, use Intel’s detection guide to check the firmware version. [1] Third, check with the system maker for updated firmware. Fourth, if no update was available, apply the measures in Intel’s mitigation guide. [1] Intel released a downloadable discovery tool to analyze a system for the vulnerability. [1] Fixed firmware carried a four-digit build number starting with 3. [1]

The advisory also listed computer makers whose update schedules Intel knew of at the time: HP, Lenovo and Fujitsu. [1] Its revision history showed the initial release and a detection update on May 1, a correction of typos on May 2 and link updates on May 3. [1]

CERT/CC published Vulnerability Note VU#491375 on May 2, 2017. [2] It listed the flaw as CVE-2017-5689, the identifier Intel’s advisory also gave, and classed the weakness as improper access control. [2] Its impact statement was blunt. A remote, unauthenticated attacker may be able to gain access to the remote management features, and execution occurs at a hardware system level regardless of operating system environment and configuration. [2]

CERT/CC flagged a gap in what anyone knew. It said it was not clear how many devices shipped with the management features enabled by default. [2] Computer makers could enable them by default on a model or BIOS/UEFI version basis, and CERT/CC was contacting makers to find out which models were affected. [2]

The makers’ answers arrived through the week, recorded in CERT/CC’s vendor entries. The entry for Lenovo, with a statement date of May 2, said some ThinkCentre, ThinkPad, ThinkServer and ThinkStation models were affected, with firmware expected in mid-May or June depending on model. [2] The entry for Hewlett Packard Enterprise, with a statement date of May 5, said some products were affected while the company assessed the total impact. [2] Dell’s statement was dated May 6. It said Dell was working on mitigation and pointed customers to Intel’s detection and mitigation guides in the meantime. [2] Dell also said there were no known exploitations reported to date. [2] Cisco, notified on May 2, said its UCS servers used Intel Server Platform Services and were not affected. [2]

An operating system patch process works in a familiar way. A tool asks each computer what software it runs, compares the answer with the vendor’s list of fixes, and pushes what is missing. The AMT flaw sat outside that loop. It lived in firmware that runs beneath the operating system, and CERT/CC said exploitation occurs regardless of operating system configuration. [2] The fix did not come from Intel to the customer. It came from Intel to each computer maker, and from each maker to its customers on that maker’s schedule. [1] [2]

Our argument is that the flaw exposed a layer that operating system patch processes do not manage, and that the gap was organizational before it was technical. To fix it, an organization had to know which machines had the feature, whether it was turned on, and which maker’s firmware applied. It then had to get, test and install an update that its usual tools were not designed to handle. Where any step failed, the organization was left with the interim measures in Intel’s mitigation guide. [1]

We take the flaw in four parts:

  1. The layer: where the flaw lived and why operating system defenses did not reach it.
  2. The inventory: what an organization had to know about each machine to decide whether it was exposed.
  3. The supply chain: how the fix traveled from Intel through each computer maker.
  4. The interim: what an organization could do while a fixed firmware image did not yet exist for its models.

The first part is about the technology. The second is about records. The third is about vendors. The fourth is about decisions under uncertainty. Together they run from the flaw itself to the day each machine was fixed or retired. After the four parts, we look at how the flaw reached different kinds of organization, answer objections in order, and list what infrastructure and security leaders could check.

A note on sources and our reading. We describe the flaw from Intel’s advisory as archived on May 6, 2017, and from CERT/CC’s vulnerability note. [1] [2] We describe later developments from Intel’s November 2017 advisory, NIST’s 2018 firmware guideline, and CISA’s 2021 directive, its 2026 replacement and its catalog. [3] [4] [5] [7] [6] Statements about how a given organization was or was not exposed are our reading, not findings in those documents.

The layer: a management engine below the operating system

The first part asks where the flaw lived. The answer explains why familiar defenses did not reach it.

What the feature does

CERT/CC described AMT and its sibling technologies as hardware-based remote management technologies for the maintenance of computer systems. [2] The Intel Management Engine supports them. [2] Remote maintenance is the point of the feature. It lets an IT team reach a fleet of machines over the network.

The same design moves the feature out of the operating system’s view. CERT/CC said the technologies listen for remote commands on known ports, and that execution occurs at a hardware system level. [2] An endpoint agent, a host firewall or an antivirus product runs inside the operating system. In our reading, none of them is placed to inspect traffic that the management engine handles before the operating system sees it.

Two paths, two scores

Intel’s advisory separated two attack paths. [1] The network path applied to provisioned AMT and ISM systems and needed no privileges. Intel scored it 9.8. [1] The local path let an unprivileged local attacker provision the features and then gain privileges. It applied to SBT as well, and Intel scored it 8.4. [1]

The distinction matters for triage. A provisioned machine was exposed to anyone who could reach its management ports. An unprovisioned but capable machine was exposed to someone with local access who could turn the feature on. [1] The second case is less urgent, and it never goes away on its own. Provisioning is a state, and states change when a machine is reimaged, repaired or handed to a new user.

Why the operating system patch did not help

An operating system update process tracks the operating system’s components. The fixed firmware for this flaw was identified by a build number in the management firmware, with resolved versions carrying a build number starting with 3. [1] An organization’s usual report of missing patches would not, by default, show that number.

In our reading, the flaw also tested the risk register. A register can list operating systems, applications and network devices as asset classes with owners, and still have no owner for the firmware of end-user computers. In that case, the question of who should act on INTEL-SA-00075 could fall between the desktop team, the security team and the procurement team that chose the hardware.

Reading the scores as an owner

The two CVSS scores told a reader how bad each path could be. [1] They did not tell an organization which path applied to which machine. That depended on the machine’s state. A 9.8 on a provisioned laptop in a branch office and an 8.4 on an unprovisioned desktop in a locked room are different risks to the organization, even though both are in the advisory.

In our reading, the useful translation of the advisory was a short decision table. Provisioned and reachable over the network: fix or contain first. Capable but unprovisioned: fix on the normal firmware cycle and make sure no one provisions the feature casually. Not capable: record the finding and close it. An organization that built that table in the first week could direct its staff. One that treated every business PC the same spent effort evenly on unequal risks.

The management server as a target

A provisioned fleet is managed from somewhere. The server that sends AMT commands holds the credentials and settings for every machine it manages. In our reading, the flaw made that server’s own protection more important, not less. If the flaw let an attacker bypass the machine’s access control, the legitimate management path still held credentials worth protecting. An organization that used AMT needed to know where its management server was, who could log in to it, and which networks could reach the managed machines.

What changed and what did not

The flaw did not change the value of operating system patching. It showed a second layer that needed the same discipline. Intel’s own later advisory showed the layer was not limited to one feature. In November 2017, Intel published a cumulative update after a security review of its Management Engine, Server Platform Services and Trusted Execution Engine. [3] We return to that in hindsight. For May 2017, the point for leaders was narrower. Their computers carried a management system they had bought, might not use, and did not patch through their normal process.

The inventory: four facts about each machine

The second part asks what an organization needed to know before it could act. Intel’s four steps implied four facts about each machine. [1]

Capability

The first fact was whether a machine was AMT, SBT or ISM capable. [1] Capability depended on the model and the firmware its maker shipped. [2] Intel said consumer PCs with consumer firmware were not affected. [1] So two machines with similar processors could differ, depending on whether they were sold as business or consumer models.

For an organization, this meant the purchase record mattered. In our reading, a fleet of business models was where to look first. A fleet assembled from mixed sources, such as computers acquired in a merger, could hold both kinds. A serial number alone was not enough. The asset register needed the model and configuration too.

Provisioning state

The second fact was whether the feature was provisioned, meaning configured for use. Intel limited the network path to provisioned AMT and ISM systems. [1] CERT/CC noted that makers could enable the features by default on some models or firmware versions. [2] An organization that had never used AMT could not assume it was off.

The state lived inside the firmware, not in the operating system that asset tools query. Finding it meant running Intel’s discovery tool or an equivalent on each machine, or querying the management interface itself. [1]

Firmware version

The third fact was the firmware version. Intel’s detection guide existed to check it. [1] The affected range ran from 6.x to 11.6, and Intel’s table gave the resolved build for each branch. [1] A machine on version 9.1, for example, needed build 9.1.41.3024 or later under the table. [1]

Collecting this fact at scale took tooling. The discovery tool analyzed a system for the vulnerability. [1] Running it across thousands of machines meant packaging it, deploying it and collecting the results into a central record. That is the same work an organization does for software inventory, applied to a layer it had not inventoried before.

Maker and model

The fourth fact was who made the machine and which model it was. Intel told customers to check with their system maker for updated firmware. [1] CERT/CC told them to check with their hardware vendor for a customized firmware update. [2] The fix was specific to the maker and model, so the inventory had to be too.

Who held each fact

The four facts could sit in different places, in our reading. Procurement held the maker and model. The desktop team held the image and perhaps the firmware version. Provisioning state might be known only to whoever had set up remote support years before. No single system joined them.

The work of the first week was therefore partly a records exercise. Someone had to pull the purchase data, run the discovery tool, and match the results to the maker’s lists of affected models. [1] [2] The output was a list of machines with a status for each. That list became the working document for the rest of the response.

The evidence an auditor would ask for

An auditor reviewing the response later would ask for the same list. Which machines were in scope, how was scope decided, which were fixed and when, and what happened to the rest. In our reading, an organization that kept the discovery tool’s output, the maker’s advisories it relied on, and its own deployment records could answer in an afternoon. One that kept none of them had to reconstruct the story from memory.

What an incomplete inventory cost

An organization missing any of the four facts had two poor choices. It could treat every machine as exposed, which meant applying mitigations or updates fleet-wide, at a cost in time and disruption. Or it could treat unknown machines as safe, which left an exposure it could not size. In our reading, the cost of the flaw for a given organization tracked the quality of its hardware records more closely than the size of its fleet.

The inventory also had a time dimension. A capable machine reimaged next month might come back provisioned. A new batch of computers might arrive with an older firmware build. The four facts had to be checked at intake and kept current, not gathered once.

The supply chain: one fix, many delivery schedules

The third part asks how the fix reached the machine. Intel released updated firmware for all affected hardware generations. [2] Intel’s release did not end the work for customers. It started a second round of work at each computer maker.

Why the maker was in the middle

CERT/CC said makers could enable the features by model or BIOS/UEFI version, and told customers to check with their hardware vendor for a customized firmware update for their product. [2] Intel’s advisory likewise pointed customers to the maker. [1]

That structure meant the date a customer could fix a machine depended on the maker’s engineering queue for that model. Intel’s release date was the earliest possible date, not the expected one.

What the makers said, and when

CERT/CC’s note recorded each maker’s status. [2] Its entry for Lenovo said firmware was expected in mid-May or June depending on model. [2] Its entry for Dell said updated firmware would begin being released on May 17, 2017, depending on the product. [2] Its entry for Toshiba said patches for some systems were available, with all affected systems expected to receive an update by July 2017. [2] Fujitsu released an advisory and updated firmware for its products. [2]

Industrial equipment followed a different path. CERT/CC recorded that it notified Siemens on May 22, 2017. [2] Siemens said that several of its industrial products use Intel technology and were affected, and that it had released updates for various products and recommended mitigations until fixes were available. [2] Its statement was dated June 26, 2017. [2]

Some appliance makers said their products were unaffected. Check Point said the built-in firmware on its appliances did not have these Intel management features. [2] Cisco said its UCS servers used Server Platform Services rather than AMT, ISM or SBT. [2] Those statements still mattered. An organization with Intel inside an appliance needed the appliance maker’s answer, because it could not inspect the firmware itself.

What a multi-maker fleet faced

A fleet with machines from several makers faced several schedules at once. Each maker published its own list of affected models, its own files and its own installation method. [2] Each update needed testing on the organization’s standard image before broad release, because a firmware update that fails can leave a machine unusable.

In our reading, firmware updates also carried more operational risk than routine software updates. A firmware update can need a restart, physical presence, or a power supply that does not fail midway. Laptops away from the office had to be reached remotely or recalled. Each of those steps took staff time that a routine operating system cycle does not.

Watching for the maker’s release

Each maker’s status changed over time. CERT/CC’s note recorded those changes, and the note itself was last revised on December 21, 2017. [2] For a customer, that meant the response could not be a single action. Someone had to watch each maker’s advisory page, notice when a model’s firmware appeared, and move that model from the interim group to the update group.

In our reading, this was where responses stalled. The first week had energy and attention. By the time a maker shipped firmware for an older model weeks later, the team had moved on. A named owner and a recurring check, tied to the list of machines from the inventory part, kept the response alive until the last model was done.

Testing before release

A firmware image that works on the maker’s test bench may still conflict with an organization’s own image, encryption software or docking hardware. A cautious team would install the update on a small pilot group first, confirm the machines booted and the firmware version changed, and then widen the release. Each stage needed a rollback plan. For firmware, rollback may be harder than for software, so the pilot group mattered more.

Contracts and procurement

The supply chain also ran through contracts. A hardware purchase agreement could say how long the maker would supply security updates for a model, how fast, and how it would notify the customer. An agreement silent on firmware left the buyer to wait. The flaw gave procurement teams a concrete reason to ask for those terms.

NIST later framed the same idea as guidance. Its 2018 platform firmware guideline said system administrators, security professionals and users can use it to guide procurement strategies and priorities for future systems. [4] We return to it in hindsight.

What a buyer could ask a maker

The flaw suggests a short set of requests a buyer can put to any hardware maker, before purchase or at renewal. In our reading, each one maps to a delay customers faced in May 2017. [2]

  • A list of the management and firmware features each model ships with, and their default state.
  • A stated period of firmware security support for each model, from the date of sale.
  • A target time from a chip supplier’s fix to the maker’s release for supported models.
  • A single page or feed where the maker posts security advisories by model.
  • A method to update firmware remotely and in bulk, with a documented rollback.

These terms would not have prevented the flaw. Each would have shortened the time between Intel’s release and a customer’s fix, or made that time predictable. Predictability matters because an organization can plan an interim control for a known period. It cannot plan well for an open-ended one.

The interim: living with exposure until a fix arrived

The fourth part asks what an organization could do before its maker shipped a fix. Intel’s answer was its mitigation guide, for affected customers without an update from their maker. [1] [2] Dell recommended the same guides in the meantime, plus best practices for securing internal networks and protecting systems from unauthorized physical access. [2]

The disable option

The simplest interim choice, in our reading, was to switch the feature off where it was not used. An organization that had never adopted AMT gave up nothing by unprovisioning it. An organization that relied on AMT for remote support faced a real trade. Turning it off would reduce exposure and reduce the help desk’s ability to fix machines remotely.

That choice belonged to someone with authority over both security and service levels. A security team acting alone could break a support process. A desktop team acting alone could leave a critical exposure open to keep a convenience. The flaw tested whether the organization had a way to make that decision quickly.

Network containment

CERT/CC’s list of ports gave network teams something to act on. [2] In our reading, an organization could restrict traffic to those ports to the management servers that needed it, and monitor for traffic from anywhere else. That reduced the network path’s reach without touching the machines. It did nothing about the local path, which Intel described separately. [1]

Containment also depended on knowing the network. A flat internal network, where any workstation could reach any other, gave an attacker inside the perimeter a direct line to every provisioned machine. A segmented network limited the reach. The flaw rewarded segmentation decisions made long before May 2017.

Exception records

Some machines could not be fixed quickly. The maker had not shipped firmware, or the machine ran a process that could not be interrupted. For those, the organization needed an exception: a record of the machine, the reason, the interim control and the date for review. In our reading, that record is what an auditor or examiner would ask for later. It shows the organization knew of the exposure and chose how to carry it.

Measuring the interim

An interim measure is only useful if someone can tell whether it is working. In our reading, an organization could track three numbers each week: provisioned machines still unfixed, capable machines still unfixed, and machines in exception. A falling first number showed the urgent risk shrinking. A flat third number showed exceptions turning into permanent arrangements. The numbers also gave leadership a way to see progress without reading the technical detail.

Telling users and leadership

The interim also had a communication side. Users whose machines needed a restart for firmware had to be told when and why. Leadership needed to know the size of the exposure and the expected end date. In our reading, a short weekly note that gave the three numbers above, the makers still pending and any decision needed would serve both purposes. It also created a dated record of what leadership knew and when, which matters if an incident later occurs on an unfixed machine.

A decision to keep the feature on for remote support was a decision to accept risk. That decision belonged in writing, with a name attached. A help desk manager could recommend it. Someone accountable for risk had to own it.

Retirement

Some machines would never get a fix. A maker could decide an old model was beyond support. For those machines the interim became permanent, and the choice was retirement or isolation. The flaw turned the age of a fleet into a security question, which in turn made hardware refresh budgets part of the security plan.

How the flaw reached different organizations

The four parts applied everywhere, and they played out differently by sector. The examples below are our reading of how the structure fit each setting.

Banks and credit unions

A bank that buys business models for its branches and back offices is buying where AMT capability was found. [1] A bank with a remote support model could have provisioned the feature on purpose. Their trade-off between remote support and exposure was sharpest. A credit union with outsourced desktop support depended on its provider to find, patch and report each machine. The contract’s reporting terms decided whether the credit union could show an examiner what had happened.

For both, the record of the response mattered as much as the response. A bank that could produce the list of affected machines, the maker advisories it relied on, the dates of each fix and the exceptions it approved would have a clear account for its board and its examiners. In our reading, the same list also fed the bank’s vendor management file, because the hardware maker had become a supplier whose security performance the bank had now observed.

Broker-dealers and wealth managers

Trading desks and advisory offices run on workstations that are hard to take down during market hours. A firmware update with a restart had to be scheduled around trading. In our reading, these firms had a strong reason to use network containment while they scheduled updates, because a failed update during the day carried a business cost.

Health systems and clinics

Hospitals run business PCs alongside equipment from other makers. Siemens said several of its industrial products use Intel technology and were affected. [2] That showed the exposure could sit inside equipment a customer does not manage itself. For any such equipment in a hospital, the hospital depended on the equipment maker’s advisory, and its biomedical or facilities engineering team had to be part of the response.

Federal agencies

An agency with a large, standardized fleet had an easier inventory problem in principle. Long hardware lifecycles and centralized buying cut the other way. A standard image deployed across an agency could carry the same firmware build on thousands of machines. We found no 2017 federal directive aimed at this flaw. The catalog that later listed it, and gave agencies a due date, came in 2021, as hindsight shows. [5]

State and local governments

A county or city that keeps machines for years, with a small IT team, faced the supply chain part as its hardest. They needed the maker’s update for each model and the staff to install it. A municipality with several generations of desktops faced several makers and several schedules with a handful of technicians.

Appliances and embedded systems

Everywhere, appliances raised the same question. A security appliance, a kiosk or a building controller may run on an Intel platform. The owner cannot inspect its firmware. Check Point’s and Cisco’s statements show the answer an owner needed: a maker’s written statement of whether the product was affected. [2] Organizations that kept those statements on file could close the question. Organizations that did not had to ask again.

Managed service providers

A provider that manages desktops for many clients faced the flaw many times over. It needed the four facts for each client’s fleet, and each client’s permission to act. Its response quality depended on how well it kept per-client hardware records. For the clients, the question was whether their contract required the provider to report affected machines and fixes. Without that term, a client could not show its own regulator what had been done on its behalf.

Objections, in order

1. Only provisioned machines were exposed over the network, so the risk was small

Intel limited the network path to provisioned AMT and ISM systems. [1] That narrowed the most severe path. It did not remove the local path, which applied to unprovisioned capable systems too. [1] And CERT/CC said makers could enable the features by default on some models. [2] An organization could not rely on the narrowing without checking its own machines.

2. Dell reported no known exploitation, so urgency was overstated

Dell said there were no known exploitations reported to date. [2] That was true when said. CISA later added the flaw to its known exploited vulnerabilities catalog, which it does on reliable evidence of exploitation. [5] [6] Our argument does not rest on exploitation in May 2017. It rests on the fact that organizations needed the same four facts and the same supply chain work whether or not the flaw was being used.

3. This was Intel’s problem to fix

Intel did fix it, for all affected hardware generations. [2] The flaw was still the customer’s problem to close, because the fix reached a machine only through its maker and the customer’s own deployment. [1] [2] Responsibility for writing the fix and responsibility for installing it sat with different parties.

4. Endpoint security tools would detect abuse

CERT/CC said execution occurred at a hardware system level regardless of operating system environment and configuration. [2] Tools that run inside the operating system have limited sight of that layer, in our reading. Network monitoring had a better chance, which is one reason containment mattered.

5. Consumer machines were unaffected, so small offices were safe

Intel said consumer PCs with consumer firmware were not affected. [1] A small office that bought business models was outside that exclusion. Intel Small Business Technology was within scope of the local path. [1] The consumer exclusion did not cover a small office’s business machines.

6. Firmware updates are too risky to rush

The objection is fair. A failed firmware update can disable a machine. That is why testing and staging mattered. It is also why the interim part of our structure exists. An organization that could not update quickly needed containment and exception records, not inaction. [1] [2]

7. The CVSS score told organizations what to do

The scores ranked the two paths. [1] They did not say which machines were provisioned, which makers had shipped firmware, or which machines could be contained. The four parts of our argument are the work the score did not do.

8. Replacing old hardware is the only lasting answer

Retirement solves the problem for a machine that will never get a fix. For machines under support, the maker’s firmware is the fix. [1] [2] Replacing whole fleets for one flaw would cost more and take longer than updating them. Replacement belongs in the refresh plan, informed by which models lose support first.

9. The flaw was a one-off

Intel’s November 2017 review found further flaws in the same management layer, including in Server Platform Services, which the May advisory had placed out of scope. [1] [3] An organization that built the four facts into its records in May was ready for November. One that treated May as a one-off had to start again.

What infrastructure and security leaders could check

The four parts translate into checks any organization can run on the firmware layer today, for this flaw or the next one in the same class.

  1. Can the asset register say, for each computer, its maker, model, firmware version and whether remote management is provisioned?
  2. Who owns the decision to turn a management feature off when it is exposed, and how quickly can they make it?
  3. For each hardware maker, does the contract state how long and how quickly firmware security updates will be supplied?
  4. Are management ports reachable only from the systems that need them, and is traffic from elsewhere detected?
  5. Does every unfixed machine have a written exception with an interim control and a review date?

The checks do not depend on Intel or on any one maker. They apply to any feature that runs beneath the operating system and is updated by the hardware maker: baseboard controllers on servers, firmware in network cards, embedded computers inside equipment. Each of those has the same four parts: a layer, an inventory, a supply chain and an interim.

The first check carries the others. Without the four facts per machine, an organization cannot decide where to disable, what to contain or which maker to chase. [1] [2] The flaw of May 2017 made that record a security control.

With hindsight

The AMT flaw turned out to be the start of a sequence. On November 20, 2017, Intel published INTEL-SA-00086. [3] It said that, in response to issues identified by external researchers, Intel had performed an in-depth security review of its Management Engine, Server Platform Services and Trusted Execution Engine. [3] The review found several vulnerabilities. Affected systems included those using Management Engine firmware versions from 6.x through 11.20, Server Platform Services 4.0 and Trusted Execution Engine 3.0. [3]

The November advisory reached further than the May one. Server Platform Services, which the May advisory had placed out of scope, now appeared in the list. [1] [3] Affected products ran from 1st through 8th Generation Core processors, plus several Xeon, Pentium, Celeron and Atom families. [3] Intel said a successful attacker could load and execute arbitrary code outside the visibility of the user and operating system. [3] Intel said the listed flaws required local or physical access, with one exception potentially exploitable over a network. [3] The fix route was the same as in May: customers were told to check with their system manufacturer for updated firmware. [3]

NIST published SP 800-193, Platform Firmware Resiliency Guidelines, in May 2018. [4] Its abstract described technical guidelines to protect platform firmware against unauthorized changes, detect changes that occur and recover rapidly and securely. [4] It addressed implementers, including original equipment manufacturers. It also said administrators and users can use it to guide procurement strategies for future systems. [4] That last point matched the supply chain part of our argument. In our reading, a buyer has the strongest say over firmware support at the point of purchase.

The federal response to exploited flaws changed in 2021. CISA issued Binding Operational Directive 22-01 on November 3, 2021. [5] It established a catalog of known exploited vulnerabilities and required federal agencies to remediate listed flaws on set timelines. [5] The directive applied to all software and hardware on federal information systems, whether on agency premises or hosted by third parties. [5] For flaws with a CVE assigned before 2021, the default timeline was six months. [5]

CISA added CVE-2017-5689 to the catalog on January 28, 2022, with a due date of July 28, 2022. [6] The required action was to apply updates per vendor instructions. [6] CISA’s guidance on the directive said it adds a flaw when there is reliable evidence of exploitation, and that adding an older flaw does not mean CISA is observing current exploitation. [5] The guidance also said that an asset that could not be updated in time had to be removed from the agency network. [5]

BOD 22-01 is no longer in force. On June 10, 2026, CISA issued Binding Operational Directive 26-04, Prioritizing Security Updates Based on Risk, which revoked and superseded BOD 22-01 and the earlier BOD 19-02. [7] [5] The new directive keeps the catalog. It sets how urgently agencies must remediate a flaw from four variables: whether the asset is publicly exposed, whether the flaw is in the catalog, whether exploitation can be automated, and whether an attacker gains partial or total control. [7] It also requires agencies to tag assets reachable from outside the agency network. [7]

The catalog entry tied our four parts together for federal agencies. To meet the 2022 due date, an agency needed to know which machines had the flaw, get the maker’s fix and install it, or remove the machine. [5] [6] That is the inventory, supply chain and interim work of May 2017, with a deadline attached.

What held up? The structure of the problem did. Both Intel advisories sent customers to their system maker for the fix. [1] [3] The layer remained below the operating system, and the fix still depended on each maker. [3] What changed was the expectation of evidence. A federal agency answered to a due date and a removal rule under BOD 22-01, and now answers to risk-based timelines and asset tagging under BOD 26-04. [5] [7] Private organizations carry no such directive, though CISA encouraged all stakeholders to use the catalog. [5] In our reading, the evidence a federal agency now keeps is the evidence any organization would want if asked which machines were affected and what happened to each.

What did not hold up was the hope that a feature an organization did not use could be ignored. The May 2017 advisory showed that unused features still counted when they were capable and could be provisioned. [1] The November review showed that the management layer itself carried risk beyond one feature. [3]

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

  • Intel AMT
  • firmware
  • CVE-2017-5689
  • patch management
  • asset inventory
  • known exploited vulnerabilities