Ransomware, IR & Endpoint

Is a Six-Digit BitLocker PIN Still Enough? TPM Lockout Rollback and CVE-2026-43613

Dark cyberpunk illustration of a brass ratchet wheel glowing amber on a circuit-board wall, with a thin blade slipped under its pawl so the wheel slips backwards, cyan filaments trailing away into the dark.

The counter that limits PIN guessing on a modern Windows laptop sits on an eight-pin flash chip, two screws from the outside world. Firmware TPMs carry no storage of their own, so they encrypt their state, including the count of failed authorization attempts, and write it to the SPI flash the platform boots from. Six researchers at Technische Universität Berlin worked out how to put an old copy of that state back. Their paper, Hello Again: Exploiting Replay Protected Monotonic Counters to Brute-Force Windows Hello PINs, landed on arXiv on November 13 and is due at ACM CCS 2026.

Rolling the state back hands back a fresh set of guesses. A four-digit Windows Hello PIN fell in 236 attempts over 57 minutes on an HP OmniBook 3. A six-digit BitLocker startup PIN fell in 236 attempts over 6 minutes 55 seconds on a Framework Laptop 13. AMD confirmed one of the two flaws on November 17 as AMD-SN-7079, CVE-2026-43613, and is shipping firmware from this month through February.

This is a device-theft problem and nothing else. It needs the case open, the flash chip off the board, and a purpose-built interposer. If your laptops never leave a controlled office, read the next section and skip the rest. If your fleet travels, the fix costs two digits and a policy change, and you can start it this morning.

What the TPM counts, and where it writes it down

Microsoft publishes the numbers. In TPM fundamentals it states that "the TPM is configured by Windows to lock after 32 authorization failures and to forget one authorization failure every 10 minutes." Those two values are the whole reason a short PIN is considered survivable. BitLocker hashes the PIN with SHA-256 and hands the first 160 bits to the TPM as authorization data, so every wrong guess spends an authorization failure, and the arithmetic caps an attacker near 4,400 guesses a year.

Discrete TPM chips keep that count in their own protected memory. Firmware TPMs, which ship inside AMD and Intel client silicon today, have nowhere to put it, so they seal the state and write it to the mainboard SPI flash. An attacker who can read and write that chip archives a known-good copy, burns 32 guesses, restores the copy, and starts again. Newer platforms answer that with Replay Protected Monotonic Counters: hardware counters inside the flash chip that only increase, authenticated with an HMAC key the chip holds.

The paper breaks RPMC in two independent ways.

  • The AMD path. The AMD Secure Processor never checks that the counter it asked to increment actually moved. AMD's bulletin calls it "insufficient authentication of RPMC Increment completion in the ASP firmware." An interposer wired between processor and flash swallows the increment and reports success, and the platform accepts the stale state. AMD scores it CVSS 4.0 5.1, CVSS:4.0/AV:P/AC:L/AT:N/PR:N/UI:N/VC:N/VI:H/VA:N/SC:N/SI:N/SA:N, CWE-345.
  • The flash path. A brief voltage glitch makes Winbond RPMC flash disclose the HMAC key that signs its counter values, in about a second under the best settings the authors found. With the key, any counter value can be forged on any processor. The BitLocker result came off an Intel machine for that reason.
  • The path with no protection at all. On two AMD desktops without RPMC-capable flash, writing an old image back with an ordinary programmer reset the lockout counters. Three Intel systems caught the same write-back.

Windows helps. The Hello test laptop forced a restart every five wrong attempts and applied no delay, and the BitLocker pre-boot prompt enforces no limit, so each restored state delivered another 32 tries.

The numbers worth planning against

Treat the headline timings as a measurement of guess rate rather than keyspace. Both recovered PINs sat early in the order the authors' scripts walked, and each worst case below was extrapolated from one laptop. The comparison that matters runs between the documented lockout and the attacked lockout, for numeric PINs:

  • Four digits. Documented worst case 69 days. Measured 40 hours against Hello on the AMD laptop, an estimated 5 hours against BitLocker on the Intel one.
  • Six digits. Documented worst case 19 years. Estimated 168 days against Hello, 20 days against BitLocker.
  • Eight digits. Documented worst case 1,903 years. Estimated 46 years against Hello, 6 years against BitLocker.

Six digits has been the Windows default for a BitLocker startup PIN since Windows 10 version 1703, so it is what most fleets run, and the number beside it moved from nineteen years to twenty days. Eight digits puts it back out of reach on the same hardware.

Who this reaches, and who can close the tab

In both demonstrations the flash chip came off the mainboard and a researcher-built board went in its place. The glitching method needs one of their PCBs, flash data extracted with psptool, and glitch timing tuned per setup; the authors say plainly that reproduction is not plug-and-play. There is no network path, no malware, no remote trigger, and no report of this outside a lab.

The scope is physical custody: laptops in checked baggage, machines handed to a repair channel, devices lost at a conference, kit recycled without a wipe. For a law firm or a clinic the consequence is the lost-laptop notification letter, and this work removes "the PIN would have held" from the assessment behind it.

You can stop here if your BitLocker policy is TPM-only with no PIN. There is nothing to guess, and Microsoft documents that a TPM-only configuration releases its key even while the TPM is locked. That configuration carries its own problem, which we covered when Microsoft shipped the YellowKey recovery-environment bypass mitigation and recommended moving to TPM+PIN. The PIN remains the better configuration, at a length above the default.

Act on this if your devices travel and your BitLocker minimum PIN length is four or six, if you run Windows Hello for Business with a four-digit PIN, or if you buy AMD client hardware, which is exposed through the interposer path with RPMC flash and through plain reflashing without it. The vendor positions matter as you plan. AMD says physical-access attacks fall outside its threat model, then says RPMC is intended to stop this class of attack and ships the fix anyway. Intel's announcement of November 16 states the work did not demonstrate a vulnerability in an Intel component and points customers at their system manufacturer. Microsoft rated the Hello report moderate and committed to recommending six digits.

Count the digits across the fleet first

Before changing policy, find out what you have. Run this elevated through your RMM or an Intune remediation script and collect it centrally.

# TPM type, BitLocker protectors, and the PIN policy actually applied on this box
$tpm = Get-Tpm
$man = (Get-CimInstance -Namespace root\cimv2\security\microsofttpm `
                        -ClassName Win32_Tpm).ManufacturerIdTxt
$bde = Get-BitLockerVolume -MountPoint $env:SystemDrive
$fve = 'HKLM:\SOFTWARE\Policies\Microsoft\FVE'
$pfw = 'HKLM:\SOFTWARE\Policies\Microsoft\PassportForWork\PINComplexity'

[pscustomobject]@{
  Computer     = $env:COMPUTERNAME
  TpmPresent   = $tpm.TpmPresent
  TpmVendor    = $man          # AMD / INTC = firmware TPM, state lives on SPI flash
  Protectors   = ($bde.KeyProtector.KeyProtectorType | Sort-Object -Unique) -join ','
  MinimumPin   = (Get-ItemProperty $fve -Name MinimumPIN     -EA 0).MinimumPIN
  EnhancedPin  = (Get-ItemProperty $fve -Name UseEnhancedPin -EA 0).UseEnhancedPin
  HelloMinPin  = (Get-ItemProperty $pfw -Name MinimumPINLength -EA 0).MinimumPINLength
} | Export-Csv -Append -NoTypeInformation \\share\audit\pin-posture.csv

Reading the output

  • Protectors shows Tpm only. No startup PIN exists, so this attack has nothing to brute force there.
  • Protectors shows TpmPin, MinimumPin blank. No policy applied, so the six-digit default stands. That is the twenty-day row.
  • MinimumPin reads 4. Somebody lowered it deliberately. Fix that device first.
  • EnhancedPin reads 1. PINs there may contain letters and punctuation, which grows the keyspace far faster than extra digits.
  • TpmVendor reads AMD or INTC. Firmware TPM, so the lockout state is on the flash chip. A discrete part reports its own maker, such as IFX or STM.

Adding digits is free; making the change stick is the work

Raise Configure minimum PIN length for startup under Computer Configuration, Administrative Templates, Windows Components, BitLocker Drive Encryption, Operating System Drives. The BitLocker FAQ confirms it accepts 4 to 20 digits. Set it to 8.

The policy governs PIN creation. Encrypted devices keep the PIN they were given until somebody changes it, so a GPO edit alone buys nothing on existing hardware. Drive the change out:

:: Confirm what is protecting this volume
manage-bde.exe -protectors -get %systemdrive%

:: Prompt the signed-in user to set a new, longer PIN
manage-bde.exe -changepin %systemdrive%

:: Move a TPM-only device to TPM+PIN (escrow the recovery key before running this)
manage-bde.exe -protectors -delete %systemdrive% -type tpm
manage-bde.exe -protectors -add   %systemdrive% -tpmandpin

:: Verify the recovery password is backed up where you can reach it
manage-bde.exe -protectors -get %systemdrive% -type recoverypassword

Two details catch people. Allow enhanced PINs for startup has to be enabled before the PIN is added to the drive, so turning it on afterwards changes nothing for devices already in the field. And Microsoft's FAQ advises setting PIN length "based on the risk tolerance and the hardware anti-hammering capabilities available to the TPMs on the computers." The TU Berlin work revises that hardware capability downward, so the length moves with it.

For Windows Hello for Business, raise Minimum PIN length under PIN Complexity to 8 on AMD hardware. A Windows 11 preview build already makes six the minimum for personal-account PINs, which shows where the default is heading.

Firmware reaches you through your OEM, not through AMD

AMD-SN-7079 lists mitigations by processor family with target dates from November 2026 to February 2027. Read the version strings carefully: ComboAM5PI_1.2.0.3n and StrixKrackanPI-FP8_1.1.0.0j are platform initialization blobs, not downloads. What you install is an OEM BIOS update built on top. Ryzen and Ryzen AI client parts land first; embedded families carry "Version TBD" against February 2027.

The researchers published an exposure check alongside the paper artifacts. It runs from a Linux live USB and reads the kernel's view of the AMD crypto coprocessor.

# AMD systems: is RPMC flash present, and is it in use?
grep . /sys/bus/pci/drivers/ccp/*/rpmc_*
# look for rpmc_spirom_available and rpmc_production_enabled

# Second opinion, where fwupd is available
fwupdmgr security

Missing files do not prove RPMC is absent, so treat a silent result as unknown rather than safe. The Winbond path has no software check: identifying the flash part means opening the case and reading the markings, typically 25R or 74M. The repository holds board designs, firmware and traces, not a working exploit.

The bulletin leaves two things open. It says nothing about AMD systems without RPMC-capable flash, where an ordinary programmer was sufficient, and it does not state whether the fixed firmware refuses replacement by an older build. Rollback protection that can itself be rolled back is worth asking your OEM about directly.

Raise the minimum PIN length, then force the reset

Pull the audit this week and sort the fleet by whether the device leaves a building. For every travelling laptop: escrow the recovery key, set the BitLocker minimum PIN length to 8, enable enhanced PINs for anything you re-enrol, and push manage-bde -changepin so people set the longer PIN instead of keeping the six digits they have. Raise the Hello for Business minimum alongside it on AMD hardware, and ask your vendor when the AGESA version in AMD-SN-7079 reaches your model. Then change one assumption in your incident process: a device that spent a night outside your custody no longer returns to the fleet on the strength of its PIN. If you want a second pair of eyes on your device-encryption posture, that is a short conversation.

Not sure what a lost laptop would actually expose?

We review device encryption posture end to end: protector configuration, PIN policy, recovery-key escrow, and what a device outside your custody really gives up. Book a session to walk through your fleet.