An open laptop motherboard sits on a repair bench, with a screen showing a stylized password recovery interface.
A four-digit Windows Hello PIN is only safe because the TPM limits how many wrong guesses an attacker gets. Researchers at Technische Universität Berlin have shown that on some PCs, someone with physical access and some hardware can undo those failed attempts. In two lab demonstrations, they recovered a Windows Hello test PIN and a BitLocker startup test PIN by repeatedly rolling back the failed-attempt state that triggers the lockout. They did not break the encryption. They undid the record of wrong guesses.

That is not a remote attack, and it does not affect every PC. Still, the work has produced an AMD CVE, a statement from Intel, and a practical question for anyone who manages laptops: how much protection is that short PIN really giving you?

An open laptop motherboard sits on a repair bench, with a screen showing a stylized password recovery interface. The two lab tests​

The research paper, Hello Again: Exploiting Replay Protected Monotonic Counters to Brute-Force Windows Hello PINs, was posted to arXiv on November 13, 2026. The authors are Christian Werling, Matti Finder, Hans Niklas Jacob, Julian Prein, Niclas Kühnapfel and Jean-Pierre Seifert. The arXiv record describes it as a slightly updated version of a paper published at ACM CCS '26.

The two demonstrations tested different protections on different hardware.

Protection testedLaptop and platformTest PINAttemptsDemonstration time
Windows HelloHP OmniBook 3, AMD Ryzen AI 5 330Four digits23657 minutes
BitLocker TPM+PINFramework Laptop 13, Intel Core Ultra 5 125H, Winbond flashSix digits2366 minutes 55 seconds

The Hello test was run against a locked machine. According to The Hacker News, the researchers attacked a locked HP OmniBook 3 with an AMD Ryzen AI 5 330 processor and a BitLocker-encrypted drive. A writeup on LaVX News explains why it took exactly 236 attempts: the setup typed PINs from 1000 upward and reached the correct PIN, 1235, after 236 tries in 57 minutes. Windows did not simply allow unlimited guessing. Windows still forced a restart after every five incorrect PIN attempts on that laptop, but no lockout delay was applied.

The BitLocker test on the Framework went after the TPM+PIN pre-boot protector. The paper says the optimized run took 6 minutes 55 seconds. An earlier run, before the attack loop was tuned, took 8 minutes 40 seconds.

Keep the numbers in proportion: 236 attempts is not a general benchmark. The researchers started counting near the right answer. The figures that matter for planning are the paper's worst-case estimates, covered below.

How undoing the lockout works​

A TPM has "anti-hammering" protection. It counts failed authorization attempts and adds lockout delays as the count rises. That is the only reason Microsoft can safely allow short numeric PINs.

The weak point is where the count is stored. The paper's abstract says most modern x86 PCs use firmware TPMs (fTPMs) that save their non-volatile state, including the lockout counters, to SPI flash on the motherboard. On newer platforms, preventing rollback of that state is left to the flash chip itself through Replay Protected Monotonic Counters (RPMC). RPMC is a JEDEC-standardized mechanism, and it runs over a bus that someone with the case open can physically reach.

The idea behind RPMC is simple: the counter only ever goes up, so an old copy of the TPM state should be detectably out of date. If an attacker can make the system accept a stale state, the failed guesses disappear after each reboot and the lockout never builds up. It's like a bouncer counting rejected IDs on a whiteboard while someone else keeps wiping it.

The researchers found two separate ways to make that happen.

  1. AMD did not verify counter increments. The firmware asked the flash chip to increment the RPMC counter but never checked that it actually went up. A cheap interposer board on the SPI bus could block the increment while the system carried on as if it had worked. This is the method used in the HP OmniBook Hello test.
  2. Winbond flash leaked its key. Voltage glitching a Winbond SPI flash chip exposed the HMAC key it uses to authenticate counter values. With that key, the interposer could fake counter responses and restore earlier state. This is the method used in the Framework BitLocker test.

The researchers note that pairing each method with Hello or BitLocker was a choice for the demonstrations. Either rollback method undermines the same lockout mechanism. The abstract says both flaws enable rollback of fTPM state whether the fTPM is AMD, Intel or Pluton.

There is a third finding that may matter more for some AMD owners. Per the Cybersecurity News summary, on two AMD desktops, writing an old flash image back reset the fTPM's lockout counters, and no special board was needed. The Hacker News adds a caveat: the researchers conclude that AMD processors likely have no other protection against this, though two desktops may not be representative of all AMD systems.

In short: the attack goes after where the lockout counter is stored, not the cryptography. RPMC was supposed to stop rollback. In the tested setups, either AMD's firmware or Winbond's flash chip let it through.

The worst-case numbers​

The paper models full-search worst cases under the rollback attack:

  • Four-digit Windows Hello PIN: about 40 hours, compared with about 69 days under the paper's model of a default 10-minute TPM lockout.
  • Six-digit BitLocker PIN (the default length): about 20 days worst case, compared with about 19 years under the same 10-minute model.

These are modeled estimates that depend on configuration. The paper notes that the Windows lockout setting it observed on the Framework system could be two hours, which would make the baseline (attack-free) times even longer. Either way, the gap between "with lockout" and "with rollback" is the real finding. It turns a guessing attack nobody would bother with into one that is merely tedious.

Why does a Hello PIN matter beyond the login screen? The authors say a compromised PIN can unlock device-bound keys and, depending on configuration, Hello-based enterprise single sign-on. They also describe how an attacker who additionally obtained the TPM-protected user key object could go further, but they clearly separate that scenario from what they actually tested. Admins should treat it as a plausible extension, not a demonstrated attack.

What Intel says​

Intel published security announcement 2026-11-16-001 on November 16. It says the attack requires physical access, specialized equipment, and a vulnerable third-party SPI flash implementation to get around rollback protection. Intel's main point is that the research "did not demonstrate a vulnerability in an Intel component" and did not show a successful attack on Intel platforms without a flaw in a third-party SPI flash device.

That is a carefully scoped statement, and it is technically consistent with the paper: the Framework test depended on the Winbond key leak. It does not mean the tested Framework laptop was safe. It means Intel considers the problem the flash vendor's. Intel recommends that customers contact their system manufacturer about available updates.

Winbond's chips are not a niche part. The paper calls Winbond the leading NOR flash vendor, with more than 25% market share. For owners trying to work out their own exposure, The Hacker News reports that they would have to open the PC and look for a Winbond chip marked 25R or 74M. The paper says a root-cause fix for the key leak requires a change to Winbond hardware. The German IT outlet S-EDV notes that the researchers name no date for a Winbond fix.

What AMD says and when fixes are planned​

AMD has accepted a CVE. AMD has published bulletin AMD-SN-7079 for the flaw. It tracks the issue as CVE-2026-43613 and gives it a CVSS score of 5.1. AMD says the Secure Processor firmware does not properly confirm that an RPMC increase finished, allowing someone with physical access to suppress counter updates. S-EDV dates AMD's confirmation to November 17 and specifies the score as CVSS 4.0 with a physical attack vector.

AMD's position is mixed. According to BastionZero's summary, AMD says attacks needing physical access sit outside its product security threat model, while acknowledging that RPMC exists to stop this kind of interposer attack. The researchers first told the company about the flaw in February.

The bulletin lists target months by processor family and firmware branch:

AMD product rows (summary)Planned target
Ryzen 3000, 4000, 5000 and 9000 desktop; Ryzen 7020; Ryzen 7045; Ryzen AI Max 400November 2026
Ryzen 4000/5000 mobile; Ryzen 6000; Ryzen 7000 and 8000 desktop; Ryzen 7030/7035/7040/8040; Ryzen AI 300/400; Ryzen AI Max 300; Threadripper 7000/9000; several Ryzen Z-seriesDecember 2026

Some families have several firmware branches with different target months, and one product row is listed as TBD. S-EDV reports that AMD plans corrected firmware for most affected processors between November 2026 and February 2027, with manufacturer BIOS updates coming after that.

Two caveats for anyone checking that table:

  • These dates are not end-user update dates. As BastionZero puts it, these are planned dates for AMD's own firmware, not updates already shipping for PCs. AMD's bulletin leaves out PCs without RPMC flash. Your fix arrives when HP, Lenovo, Dell, Framework or your motherboard vendor ships a BIOS that includes it.
  • The CVE covers only the AMD increment flaw. It does not cover the Winbond key leak or the reflash rollback seen on AMD desktops without RPMC.

What Windows admins and owners should do​

The threat model is narrow but real: a lost or stolen laptop, an unattended device in a hotel room, or a targeted seizure. Remote attackers are not part of it. Ranked by how quickly you can act:

  1. Lengthen PINs now. Per Comss.ru, the researchers advise adding at least two digits to PINs. The worst-case math makes the reason obvious: every added digit multiplies the search space by ten, and rollback does nothing to that multiplier. This is the only step you can take today that does not wait on a vendor.
  2. Enforce it through policy. Windows Hello for Business PIN complexity can be set centrally through Intune or Group Policy, including minimum length and whether letters are allowed. BitLocker has a separate policy for minimum startup PIN length and for "enhanced" PINs that permit non-numeric characters. Check the exact setting names in your own management console. (This step comes from general Windows administration knowledge, not from the research.)
  3. Track firmware from your OEM, not AMD's table. Find your exact model on the manufacturer's support page and watch for BIOS updates that mention CVE-2026-43613. Intel recommends the same for Intel systems with vulnerable flash.
  4. Don't treat "no RPMC" as safe. The reflash finding on two AMD desktops suggests that missing RPMC can mean missing rollback protection, not immunity.
  5. Keep physical security in your risk register. These attacks need the case open and time with the hardware. Firmware updates fix the AMD logic bug, but no firmware update can stop someone from attaching a probe to the board.

The bigger picture​

For years, firmware TPMs have been sold as a cheap, built-in alternative to discrete TPM chips, and they made Windows 11's TPM requirement workable on hundreds of millions of PCs. Hello Again is a reminder that an fTPM is only as trustworthy as the flash chip holding its state and the firmware checking that chip's answers. RPMC was meant to close the rollback gap. In the tested setups, one vendor didn't verify the counter and another's chip gave up its key under voltage glitching.

The fair reading is somewhere between "Windows Hello is broken" and "nothing to see here." Short PINs were always relying on the lockout counter to keep them safe, and in these lab setups the counter could be wound back. AMD's fixes will take weeks to months to reach actual BIOS updates, and the Winbond issue has no fix date. Until then, a longer PIN is the practical defense.

 

References

  1. Researchers bypassed Windows Hello and BitLocker PIN lockouts in two lab tests NeoTeo 2026-11-19T08:29:29+00:00
  2. TPM Lockout Bypass Cracks Windows Hello, BitLocker PINs bastionzero.com
  3. Hello Again: TPM-Sperre für Hello- und BitLocker-PINs s-edv.com