BlueBorne put the Bluetooth fix in the hands of whoever shipped each device’s software

On September 12, 2017, CERT/CC published a note on BlueBorne, eight Bluetooth flaws across Windows, iOS, Linux and Android. In our reading, each device’s fix depended on who controlled its Bluetooth software, which turned a radio into an inventory question.

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 September 12, 2017, CERT/CC published a note on BlueBorne, eight Bluetooth flaws across Windows, iOS, Linux and Android. In our reading, each device’s fix depended on who controlled its Bluetooth software, which turned a radio into an inventory question.

  1. The week On Tuesday, September 12, 2017, the CERT Coordination Center at Carnegie Mellon University’s Software Engineering Institute published Vulnerability Note VU#240311.
  2. Platforms patched by their own maker: Windows and iOS For Windows, the Bluetooth stack ships with the operating system.
  3. Android: a fix written once and delivered by manufacturers and carriers Android’s four BlueBorne flaws were fixed in its own Bluetooth stack, separate from the Linux kernel’s BlueZ.
  4. Embedded Linux: a kernel fix that waited for each device vendor’s firmware Two BlueBorne flaws sat in BlueZ, the Bluetooth stack in the Linux kernel.
  5. Devices nobody would update: where only switching off the radio remained For devices with no update, the CERT/CC note offered one workaround: disable Bluetooth if it is unused or unnecessary.
  6. How the exposure reached banks, health systems and agencies Banks run laptops, phones and tablets on the first two paths.
  7. Objections, in order Range does limit attackers, and CERT/CC scored the access vector as adjacent network.
  8. What a platform owner could check The list below is a set of checks we would run today.
  9. With hindsight CERT/CC kept the BlueBorne note current for a time.

The week

On Tuesday, September 12, 2017, the CERT Coordination Center at Carnegie Mellon University’s Software Engineering Institute published Vulnerability Note VU#240311. Its title was Multiple Bluetooth implementation vulnerabilities affect many devices. The note described a collection of flaws that had been publicly released under the name BlueBorne.

The note said the flaws together affected Windows, iOS, and Linux-kernel-based operating systems including Android and Tizen. In the worst case, it said, they could allow an unauthenticated attacker to perform commands on the device. The flaws had been disclosed by researchers at the security firm Armis, whom the note credited.

CERT/CC listed eight CVE identifiers. Two concerned Linux. CVE-2017-1000251 was a stack buffer overflow in the L2CAP code of the Linux kernel’s BlueZ module, present from kernel version 3.3-rc1 onward. CVE-2017-1000250 was an out-of-bounds read in BlueZ’s implementation of the Service Discovery Protocol, in all versions.

Four concerned Android’s separate Bluetooth stack. CVE-2017-0785 was a similar SDP read flaw. CVE-2017-0781 was a heap overflow in Android’s BNEP code. CVE-2017-0782 was an integer underflow in the same component. CVE-2017-0783 was an incorrect security level in the PAN profile that could permit a man-in-the-middle attack.

One concerned Windows. CVE-2017-8628 was a PAN profile flaw like the Android one, in all versions of Windows from Vista to Windows 10. One concerned Apple. CVE-2017-14315 was a heap overflow in Apple’s Bluetooth Low Energy Audio Protocol in iOS 9.3.5 and lower and tvOS 7.2.2 and lower.

The note stated the impact plainly. An unauthenticated, remote attacker might obtain private information about the device or user, or execute arbitrary code on it. CERT/CC scored the set at 7.9 on the CVSS version 2 base scale. The access vector was adjacent network: an attacker needed to be within Bluetooth range.

The resolution section was the part that mattered for planning. It said patches were available in the latest releases of Windows, iOS, the Linux kernel and Android. It pointed readers to Microsoft’s advisory and to Android’s September 2017 security bulletin. Microsoft’s statement on the note’s vendor list said it had released security updates on July 11, 2017. It told users to check with their device manufacturer about firmware updates.

It then added a warning. Android phones and other mobile devices in the United States, the note said, were likely to see delayed updates, or possibly never receive them. It gave the reason as the complexity of the US mobile ecosystem, in which manufacturers and carriers must support an update before it reaches the device.

For devices without an update, CERT/CC offered one workaround. Affected users should consider disabling Bluetooth if it was unused or unnecessary.

Google had published the Android fixes on September 5. Its Android Security Bulletin for September 2017 was dated September 5. It rated CVE-2017-0781 and CVE-2017-0782 as critical remote code execution flaws. It listed updated AOSP versions from 4.4.4 through 8.0. The bulletin said partners had been notified at least a month before, and that Google had no reports of active customer exploitation of the new issues.

The bulletin also explained how Android fixes travel. It used two security patch level strings, 2017-09-01 and 2017-09-05, so partners could ship a common subset sooner. Source code patches went to the Android Open Source Project. A device’s patch level, not its Android version, showed whether a fix was present.

Federal guidance on Bluetooth had been updated that spring. In May 2017, the National Institute of Standards and Technology issued revision 2 of Special Publication 800-121, its Guide to Bluetooth Security. Its checklist told organizations to maintain a complete inventory of Bluetooth-enabled devices and their addresses. It told them to change insecure default settings and to test and deploy Bluetooth software and firmware patches.

So the week brought a set of flaws in code that sat below the network layer that IT teams watch with scanners. The fixes existed for some platforms on day one. For others, they depended on a manufacturer, a carrier or a firmware vendor that the device owner did not control.

Bluetooth flaws were not new in 2017. NIST’s guide listed older attacks by name, from bluesnarfing to bluebugging. BlueBorne differed in two ways that the CERT/CC note made clear. It described an unauthenticated attacker within radio range who might run code on the device, and the flaws reached the code paths of four major operating systems at once.

The second difference shaped the response. The same week, a fully patched Windows laptop and an unpatched Android phone could sit on the same desk. An infusion pump or a lobby kiosk running embedded Linux could sit nearby, with no patch at all. Each was exposed to the same class of attack. Each would be fixed by a different party, on a different schedule.

Our argument is that BlueBorne made the Bluetooth stack a platform component with an owner, and that the owner, not the device’s user, set the pace of the fix. We take the update paths as four parts, sorted by who controlled the Bluetooth software:

  1. Platforms patched by their own maker: Windows and iOS.
  2. Android: a fix written once and delivered by manufacturers and carriers.
  3. Embedded Linux: a kernel fix that waited for each device vendor’s firmware.
  4. Devices nobody would update: where only switching off the radio remained.

Every Bluetooth-capable device an organization held falls into one of these four. A device may move between them over its life, as support ends. After the four parts, we look at how the exposure reached banks, health systems and agencies, at objections, and at what a platform owner could check.

A note on sources and our reading. We describe the flaws from the CERT/CC note, Android’s September 2017 bulletin and NIST’s May 2017 guide. We describe later events from a second CERT/CC note and from the Food and Drug Administration. Statements about who controlled which fix are our reading of those documents. Organizations should rely on their vendors’ current advisories.

Platforms patched by their own maker: Windows and iOS

One vendor, one channel

For Windows, the Bluetooth stack ships with the operating system. Microsoft fixed CVE-2017-8628 in its own update and listed it in its own advisory, which the CERT/CC note cited. Its statement in the note said the updates had been released on July 11, 2017, before the note was published on September 12, 2017. An organization that already ran Windows updates centrally had the fix through the channel it used every month.

For Apple’s devices, the note described the flaw as affecting iOS 9.3.5 and lower and tvOS 7.2.2 and lower. The fix was the newer operating system. Apple controlled both the hardware and the software, so the path from fix to device ran through one company.

In our reading, these platforms posed the smallest planning problem of the four. The remaining risk lay in devices that were outside central management, or that could not move to a current release.

Older releases that could not move

The iOS flaw affected older releases. A device that could not run a newer release stayed exposed. That is a hardware support question, not a patch question. Organizations holding older tablets for kiosks, sign-in desks or training rooms had to decide whether to replace them or switch off Bluetooth.

On Windows, the flaw reached back to Vista. Some organizations still ran older Windows versions on special-purpose machines. A fix in the current release did not help a machine frozen on an older one for a vendor’s software. The question became whether that vendor allowed the update, and whether the old release still received fixes.

Fixes that need a restart

A patch installed but not active protects nothing. Operating system updates that touch the Bluetooth stack can need a restart before the new code runs. Machines that stay on for long periods, such as shared workstations and kiosks, may report the update as installed while the old code still runs.

Patch reporting should show whether the device restarted after the update. In our reading, that small detail decides whether the first path is as complete as it looks on a dashboard.

Managed and unmanaged copies

A platform patch reaches a managed device through the organization’s tools. It reaches an unmanaged one only if the user accepts it. Personal phones used for work, contractor laptops and shared devices fall into the second group.

NIST’s guide spoke to that split. It recommended a centralized security policy management approach where available, to keep device configurations compliant. It also relied on user awareness, including turning Bluetooth devices off when they are not needed.

What an organization could verify

For this first path, the evidence was ordinary. Patch reports showed which Windows machines had Microsoft’s July 2017 updates. Device management showed iOS versions. The gap was devices that never appeared in those reports.

In our reading, BlueBorne’s first lesson for platform owners was that the Bluetooth radio did not appear on an ordinary network scan. A device could be invisible to network tools and still reachable over the air. Inventory had to come from device management, purchasing records or physical surveys, not from the network.

Bluetooth as a default setting

NIST’s guide said default settings are generally not secure and should be reviewed against the organization’s policy. It gave disabling unneeded Bluetooth profiles and services as an example.

A laptop used only with a wired keyboard does not need Bluetooth at all. A phone used with a headset needs some profiles but not others. Setting those defaults centrally reduces the code that an attacker can reach. It works on the first path because the platform maker exposes the controls to the organization’s management tools.

Peripherals paired to managed machines

A managed laptop may pair with an unmanaged headset, keyboard or mouse. BlueBorne’s flaws sat in the host’s stack, so patching the laptop closed them on that side. The peripheral’s own firmware is a separate question, and its maker controls it.

That pairing also widens who can be near the laptop. Staff use wireless headsets in open offices, call centers and clinics. The headset is harmless to the BlueBorne flaws once the host is patched. Until then, a radio left on for the headset is a radio an attacker can reach.

Laptops that travel

A laptop on the first path is fixed only when it checks in. One that spends weeks in the field, at home or with a traveling examiner may miss an update cycle. Bluetooth risk travels with it into airports, hotels and client sites, where strangers come within range.

Patch reports should therefore show the age of each device’s last check-in, as well as its patch status. A device that has not reported is an unknown, not a pass. In our reading, BlueBorne made that distinction concrete for a radio that attackers could reach from a nearby seat.

Android: a fix written once and delivered by manufacturers and carriers

Four fixes, published before the disclosure

Android’s four BlueBorne flaws were fixed in its own Bluetooth stack, separate from the Linux kernel’s BlueZ. Google listed them in the September 2017 bulletin, dated September 5. Two were rated critical remote code execution flaws.

The bulletin said partners had been notified at least a month earlier. Source code went to the Android Open Source Project. On paper, the fix existed before the BlueBorne name appeared in the CERT/CC note.

The chain from code to phone

A fix in Android’s source code is not yet a fix on a phone. The device maker must merge it into its own build, test it on its hardware and publish an update. In the United States, the CERT/CC note said, carriers must also support the update before it reaches the device.

Each link in that chain has its own schedule. A manufacturer may update flagship models first and older ones later or not at all. A carrier may bundle the fix into a larger release. The device owner sees the result as a patch level date in the settings screen.

Patch level as the evidence

The bulletin defined two patch levels for that month, 2017-09-01 and 2017-09-05. A device reporting either level, or a later one, had the Bluetooth fixes listed under the first. That made the patch level a single, checkable fact.

For an organization, this was useful. Mobile device management tools can read the patch level. A policy could set a minimum patch level for access to corporate email or apps. Devices below it could be blocked or warned.

The delay CERT/CC predicted

The CERT/CC note warned that US Android devices were likely to see delayed updates, or possibly never receive them. That warning was about the chain, not the code. Google had done its part. What remained depended on parties the device owner did not choose and could not direct.

For corporate-owned phones, an organization could choose models with a known update record. For personal phones used for work, it could only set access rules. In our reading, BlueBorne made the device maker’s update policy a procurement criterion, not a detail.

Android inside other things

Android does not run only on phones. It can run on tablets used as point-of-sale terminals, check-in kiosks, rugged handhelds and carts. Those devices follow the same chain, sometimes with an extra link: the solution vendor that built its software on a particular Android build.

That vendor may not move to a new build quickly, because its own software must be retested. An organization that bought a tablet-based system bought its update chain too. The contract for that system decided whether a Bluetooth fix would arrive, and when.

What an organization could ask of its Android vendors

Three facts follow from the bulletin’s structure. The first is the patch level each device reports today. The second is how long after Google’s monthly bulletin the vendor ships an update. The third is the date until which the model will receive updates at all.

Those facts belong in purchasing, before a fleet is bought. After purchase, the answers can only be observed. A vendor that commits to a schedule in writing gives the buyer something to measure.

Corporate-owned and personal phones

An organization can choose corporate-owned phones by their update record. It can buy models with published support periods and replace them on a schedule. It controls which patch level each device must reach and can enforce it.

Personal phones used for work are different. The owner chose the model, the carrier and the update habits. The organization can only set conditions for access: a minimum patch level, a management profile, or a separate work container. Each condition limits exposure without removing it.

In our reading, BlueBorne sharpened that choice for regulated firms. A bank or health system that lets staff read sensitive email on personal phones accepts the update chain of every model in use. The access rule is the only lever, and it works only if the patch level is checked.

Measuring the chain

Two numbers describe this path. One is the share of devices at or above the required patch level. The other is the time from Google’s bulletin to the date each model’s update reaches the organization’s devices. Both can be read from management tools once collected.

The second number is the one a vendor can be held to. It measures the chain, not the device owner. Collected over time, it shows which makers and carriers deliver and which do not. That record is useful at the next purchase.

Embedded Linux: a kernel fix that waited for each device vendor’s firmware

BlueZ in the kernel

Two BlueBorne flaws sat in BlueZ, the Bluetooth stack in the Linux kernel. The CERT/CC note said the remote code execution flaw affected kernel versions from 3.3-rc1 onward. The SDP read flaw affected all versions of BlueZ.

The note said patches were available in the latest releases of the Linux kernel. For a desktop Linux distribution, that meant an ordinary package update. For a device whose software was built once by its manufacturer, it meant waiting for new firmware.

Devices built on Linux

The note named Tizen, a Linux-based system used in consumer electronics. Linux can also run inside printers, network appliances, building controls, set-top boxes and medical equipment. Each manufacturer chose a kernel version, configured it and shipped it as firmware.

The note’s advice for such devices was to check with the manufacturer. That advice describes the situation exactly. The device owner could not apply the kernel fix. Only the maker could rebuild the firmware, test it and release it.

Kernel versions as a long tail

The remote code execution flaw reached back to kernel version 3.3-rc1. Device makers choose a kernel when they design a product and may keep it for the product’s whole life. A device designed long before 2017 could carry an affected kernel with no newer version available from its maker.

That makes the kernel version a fact worth recording at purchase. A buyer who knows that a device runs an old kernel can ask how the maker will handle flaws found later. A buyer who does not know learns it only when an advisory arrives.

The same logic applies to Android builds inside appliances. The build fixes the stack version, and the stack version fixes which flaws apply. Recording both turns each new advisory into a lookup rather than a survey.

Whether Bluetooth is even compiled in

Not every Linux device includes BlueZ. Not every device with BlueZ has a Bluetooth radio switched on. The first task for an owner was to find out which devices had both. That information sat with the manufacturer, not on any network scan.

A software bill of materials would answer the question quickly. In our view, such lists were not yet a routine part of device purchases in 2017; the federal requirement for medical cyber devices came in 2023, as the hindsight below shows. An organization had to ask each vendor in turn, and wait for answers.

Testing burdens differ by device

A consumer device maker might ship firmware quickly. A medical device maker must check that a firmware change does not affect safety or performance. An industrial controller maker may need a site visit to install new firmware. Each kind of device set its own pace.

FDA’s later SweynTooth communication addressed this burden for medical devices. It said that, in general, patches made to address those Bluetooth Low Energy flaws were not likely to need premarket review. That statement came in 2020, about a different set of flaws. In September 2017, manufacturers and their customers had to work through that question for themselves.

Contracts that decide the pace

For this path, the contract with the device vendor decided the fix. It mattered whether the vendor had committed to monitor vulnerabilities in the components it used. It mattered whether it had committed to ship fixes, and how fast. It mattered whether support ended on a date the buyer knew.

NIST’s guide recommended fully testing and regularly deploying Bluetooth software and firmware patches. That recommendation assumes the patch exists. On the embedded path, whether it exists is a commercial fact. In our reading, BlueBorne turned firmware support terms into security controls.

Engineering owners and security owners

For embedded devices, the people who understand the hardware are engineers: biomedical, facilities, network or field staff. The people who track vulnerabilities sit in security. BlueBorne required both. An advisory means little to someone who does not know which devices contain the affected stack.

A working arrangement pairs them. Security watches advisories and maps them to components. Engineering confirms which devices contain those components and plans the fix. In our reading, that pairing was the practical answer to the third path, and its absence would slow any response.

The inventory problem

NIST’s checklist told organizations to maintain a complete inventory of Bluetooth-enabled devices and their Bluetooth addresses. For phones and laptops, management tools can help. For embedded devices, the inventory depends on purchasing records, asset tags and engineering staff who know what is inside each box.

A health system’s biomedical engineering team, a bank’s facilities team and an agency’s building management team each held part of that inventory. None of them necessarily reports to the security office. In our reading, BlueBorne required those inventories to be joined.

Vendor advisories as a feed

For embedded devices, the vendor’s security advisory is the only notice that a fix exists. Some makers may publish advisories on a schedule. Others may answer only when asked. An organization holding dozens of device types needs a way to track each maker’s notices.

In our reading, that tracking is a task with an owner, not a hope. Someone has to know where each maker publishes, check it, and match each notice against the inventory. Without that, a firmware fix can exist for a long time without reaching the devices that need it.

Interim measures while firmware is pending

Between the disclosure and a firmware release, an owner has choices. Bluetooth can be switched off where the function is not needed. Devices can be moved away from public areas. Pairing can be limited to known devices, where the device allows it.

These steps do not fix the flaw. Each narrows the set of people who can reach it. For a device in a staff-only room, the remaining risk may be acceptable for a time. For one at a public counter, it may not.

Devices nobody would update: where only switching off the radio remained

The workaround as the whole answer

For devices with no update, the CERT/CC note offered one workaround: disable Bluetooth if it is unused or unnecessary. For an old phone, that is a setting. For an embedded device, it may require a configuration change by the vendor, or may not be possible at all.

If Bluetooth is used, the workaround removes a function. A tablet that talks to a Bluetooth card reader, printer or barcode scanner stops working as designed. The organization then has to choose between the risk and the function.

End of support as a security state

A device that has reached the end of its maker’s support is in this fourth path for good. Its code will not change. The only controls left are around it: where it sits, who can get near it and what it connects to.

NIST’s guide included physical measures. It asked users to protect handheld devices from theft and to pair devices in a physically secure area where attackers cannot observe passkey entry or eavesdrop on the pairing. Bluetooth range limits who can attack. A device in a locked room is harder to reach than one at a public counter. In our reading, location became a control for this path.

Replacement as the remediation

Some devices could not be patched and could not do without Bluetooth. For them, the only full remediation was replacement. That turned a vulnerability into a capital decision, with budgets, procurement rules and lead times.

An organization that knew its unsupported devices in advance could plan replacements over time. One that discovered them because of BlueBorne had to compress that planning. The inventory from the third path fed directly into the fourth.

Compensating controls and their limits

Network controls do not see Bluetooth traffic. A firewall rule cannot block an attack that travels over the air between two devices. Monitoring for unusual behavior on the device itself is the remaining option, where the device allows it.

FDA’s 2020 communication on SweynTooth recommended monitoring medical devices for signs of unusual behavior where possible. The same advice applies to other unpatched devices. It is weak protection. It detects problems after they start, not before.

Retiring Bluetooth from a design

One response to repeated Bluetooth flaws is to remove the need for it. A card reader can connect by cable. A printer can sit on a wired port. A scanner can dock. Each change removes a radio from the exposure count.

Wired links have their own costs: cabling, mobility and clutter at the counter. The choice is a design trade-off, not a security rule. In our reading, it belongs in the design review of new systems, where it costs least to change.

Documenting the accepted risk

When an organization keeps an exposed device in service, it accepts a risk. In our reading, that acceptance should be written down: which device, what exposure, what controls, who decided and when it will be reviewed. Auditors and examiners may ask for that record.

Without it, the device stays exposed by default rather than by decision. The difference matters when an incident occurs. A documented acceptance shows the organization knew and weighed the risk. Its absence suggests nobody owned the question.

Isolating by location

Bluetooth’s short range makes location a real control. Moving a device from a lobby to a back office reduces the number of people who can come within range. Shielding is possible in some settings, though costly. A device kept in a locked room behind staff-only doors has a smaller exposure than one on a public desk.

Location controls need maintenance too. Devices move. A tablet set up in a back office may end up at the front desk after a reorganization. The inventory has to record where each exposed device is allowed to be, and someone has to check.

Who decides

The fourth path ends in a decision: replace, isolate, or accept. In our reading, that decision belongs to the business owner of the device, advised by security and engineering. The security office can describe the risk. It cannot decide that a clinical or branch function should stop.

An organization with named owners for its devices can make those decisions quickly. One with only an IT asset list may find that the list does not say who can decide. BlueBorne exposed that gap as much as any gap in code.

How the exposure reached banks, health systems and agencies

Banks and credit unions

Banks run laptops, phones and tablets on the first two paths. Branches may also hold tablets, kiosks and peripherals built on Android or embedded Linux. Bluetooth links between a tablet and a card reader or printer are a design choice some vendors make.

We cite no banking guidance on Bluetooth here. In our reading, BlueBorne tested whether a bank’s asset inventory reached the branch counter, or stopped at the network edge.

A bank’s vendors matter here as well. Branch equipment can be supplied and serviced by third parties. Their update terms are the bank’s exposure. We would expect third-party risk management for those vendors to cover firmware support for wireless components.

Health systems

Health systems hold every kind of device on the four paths. Clinicians carry phones. Workstations run Windows. Clinical devices run embedded systems, some with Bluetooth links to monitors, tablets or patient-worn sensors. Patients bring their own devices, some of which connect to hospital systems.

Medical devices raise the stakes of the third and fourth paths. A firmware change needs the manufacturer’s validation. Switching off Bluetooth may disable a clinical function. FDA’s 2020 communication on Bluetooth Low Energy flaws, discussed below, asked providers to find affected devices both in their facilities and among their patients.

HIPAA’s Security Rule requires covered entities and business associates to conduct an accurate and thorough assessment of risks and vulnerabilities to electronic protected health information. In our reading, a Bluetooth-capable device that stores or transmits such information falls inside that assessment, and BlueBorne was a test of whether the assessment included the radios. Covered entities should confirm the scope with counsel.

Patient-owned devices

Patients can bring phones, hearing aids, glucose monitors and other devices into hospitals and clinics. Some pair with hospital apps or tablets. The health system does not control their software. It controls only what its own systems accept from them.

That boundary matters for BlueBorne’s first two paths. A hospital tablet patched on the first or second path can still pair with a patient’s unpatched phone. The exposure then sits on the patient’s device, outside the hospital’s reach. Clear rules on what hospital equipment may pair with reduce that exposure.

Federal agencies

NIST developed its guide under its responsibilities in the Federal Information Security Modernization Act, for federal systems. Its checklist called for a complete inventory of Bluetooth-enabled devices, with addresses. BlueBorne gave agencies a reason to check whether they had one.

Agencies also hold building systems, vehicles and field equipment that may carry Bluetooth. Those assets can fall outside the CIO’s direct management. The CERT/CC advice to check with the manufacturer landed on facilities and fleet managers as much as on IT staff.

Field and vehicle equipment

Agencies deploy rugged tablets, printers and scanners in vehicles and in the field. Inspectors, police, social workers and benefit staff carry them into homes and public places. Bluetooth links between those devices can be common in mobile work.

Field equipment is hard to update. It may connect only occasionally, and it may run a vendor’s fixed software build. Those traits put it on the second or third path, with long gaps between fixes. Its users work among the public, so the range limit protects less.

State and local governments

State and local agencies run the same mix, in some cases with smaller teams. Public counters, libraries and transit systems may host kiosks or tablets in public spaces, within easy Bluetooth range of anyone. For them, the fourth path’s physical controls carry more weight.

Firms that build the devices

For device makers, BlueBorne was a test of their own supply chain. A firm that built on a kernel or Android version needed to know which version it used and whether it included the affected code. It then had to ship fixes to customers who might not know to ask.

In our reading, a maker that could answer quickly earned trust, and one that could not left its customers on the fourth path without warning. The week showed buyers which suppliers knew their own software.

Examiners, auditors and insurers

Supervisors and auditors look for evidence that an organization knows its assets and manages their vulnerabilities. BlueBorne gave them a concrete test case. The evidence was the same for every sector: an inventory that included radios, a record of fixes applied by path, and documented decisions for devices left exposed.

Any description of asset management given to an outside party, such as an insurer, faces the same test. An answer that covers networked computers alone leaves out the devices on the third and fourth paths. In our reading, the accurate answer describes all four.

Objections, in order

1. The attacker had to be within Bluetooth range, so the risk was small

Range does limit attackers, and CERT/CC scored the access vector as adjacent network. But branches, hospitals and public offices are places where strangers come within range by design. The limit reduces the number of attackers. It does not remove the exposure at public counters, waiting rooms and lobbies.

2. There were no reports of exploitation, so urgency was misplaced

Google’s bulletin said it had no reports of active customer exploitation of the new issues. That was true when published. The question for an owner was not whether attacks had happened, but how long fixes would take to arrive. For the third and fourth paths, that answer could be never.

3. Mobile device management already handles patching

Management tools handle the first two paths well, for enrolled devices. They do not reach embedded devices, unenrolled personal devices or unsupported hardware. BlueBorne’s fixes split along exactly those lines. The tools covered the part of the problem that was easiest to fix.

4. Switching Bluetooth off everywhere solves the problem

It removes the exposure, and NIST recommended turning Bluetooth off when not needed. But some functions depend on it: headsets, peripherals and some clinical links. Where it is needed, switching it off trades a security risk for an operational one. The decision needs an owner who can weigh both.

5. This was a vendor problem, not the customer’s

Vendors wrote the fixes. But customers chose the vendors, signed the support terms and decided whether to keep unsupported devices. In our reading, the customer owned the decision even where the vendor owned the code.

6. Later flaws would make this round irrelevant

Later flaws did come, as the hindsight below shows. That supports the argument rather than weakening it. Each new set of flaws followed the same paths. In our reading, an organization that had mapped those paths in 2017 would have had less to rebuild.

What a platform owner could check

The list below is a set of checks we would run today. It reflects our reading of the sources, not a regulatory requirement.

  • An inventory of Bluetooth-capable devices, including embedded and clinical devices, with the party that controls each device’s Bluetooth software.
  • For each device, which of the four paths it is on today, and when it will move to the fourth because support ends.
  • For Android devices, the reported security patch level and a minimum level enforced for access.
  • For embedded devices, written vendor terms on vulnerability monitoring, fix delivery and end of support.
  • Default Bluetooth settings on managed laptops, phones and tablets, with unneeded profiles disabled.
  • Recorded risk acceptances for exposed devices kept in service, with physical or monitoring controls and a review date.
  • A procurement step that asks suppliers for a list of software components, including the Bluetooth stack and its version.

With hindsight

CERT/CC kept the BlueBorne note current for a time. Its record shows the note was last updated on November 8, 2017. As checked on November 15, 2026, the note still lists the same eight CVE identifiers and the same advice: apply updates, check with the device manufacturer, and disable Bluetooth where it is not needed.

Bluetooth flaws kept arriving, and they reached deeper layers. On August 14, 2019, CERT/CC published VU#918987 on what it called the Key Negotiation of Bluetooth, or KNOB, attack. The note placed the flaw in the encryption key length negotiation of Bluetooth BR/EDR Core v5.1 and earlier, which the Bluetooth core specification defines.

The KNOB note said an unauthenticated, adjacent attacker could force two devices to agree on an encryption key with as little as one byte of entropy. The attacker could then brute-force the key and decrypt traffic for that session. In our reading, this added a fifth party to our four paths: the body that writes the standard.

The KNOB fix ran through that body and down the chain. CERT/CC told Bluetooth host and controller suppliers to follow the Bluetooth SIG’s Expedited Errata Correction 11838. It told downstream vendors to refer to their suppliers for updates. In our reading, a device owner was now three or four links from the fix.

The Food and Drug Administration issued a safety communication on SweynTooth, with content dated March 3, 2020. It described 12 vulnerabilities associated with Bluetooth Low Energy. It said an unauthorized user could exploit them wirelessly to crash or deadlock a device, or to bypass security and reach functions normally available only to an authorized user.

FDA said it was not aware of confirmed adverse events. It said software to exploit the flaws in certain situations was already public. It named several makers of system-on-a-chip components that were affected. In our reading, the flaws therefore sat in parts that medical device makers bought from chip suppliers.

FDA’s recommendations followed the paths we describe. Manufacturers were to assess which devices were affected, develop compensating controls while writing patches, and tell customers. Health care providers were to work with manufacturers to identify affected devices in their facilities and among their patients. Where possible, both were to monitor devices for unusual behavior.

The regulatory framework then changed. FDA’s guidance on cybersecurity in medical devices, in the version current as of November 15, 2026, dates from February 2026. It describes section 524B of the Food, Drug, and Cosmetic Act, added by the Food and Drug Omnibus Reform Act of 2022 and enacted on December 29, 2022.

The guidance says that from March 29, 2023, sponsors of premarket submissions for cyber devices must include information to show the device meets section 524B’s cybersecurity requirements. One of those requirements, under section 524B(b)(3), is a software bill of materials. That is the component list we said was missing from purchases in 2017.

The guidance also speaks to Bluetooth directly. Where common protocols such as Bluetooth and Bluetooth Low Energy enable interoperability, it says manufacturers should assess whether added security controls beneath that communication are needed. Its example is protection against risks if vulnerabilities in the protocol or supporting technology are discovered.

What held up from the week is the structure. Each later Bluetooth flaw reached devices through whoever controlled the software, and the chain grew longer, not shorter. What changed is that, for medical devices that meet the definition of a cyber device, the law as FDA’s guidance describes it now requires makers to supply a component list and a plan to monitor, identify and address postmarket vulnerabilities. For banks’ and agencies’ embedded equipment, the sources we cite show no equivalent federal requirement.

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

  • Bluetooth
  • BlueBorne
  • patch management
  • medical devices
  • mobile devices
  • NIST SP 800-121