# Hello Again: Bypassing TPM Lockout to Brute-Force Windows Hello and BitLocker PINs

## Overview

In **November 2026**, six researchers from **Technische Universität Berlin (TU Berlin)** published **"Hello Again: Exploiting Replay Protected Monotonic Counters to Brute-Force Windows Hello PINs"**, presented at **ACM CCS '26** (an updated arXiv version was posted on November 13, 2026).

The research question sounds simple: **why is a PIN of only four digits considered secure?** Microsoft's answer has always been: because **the TPM limits wrong guesses**. Per Microsoft's documentation, the TPM locks after **32 wrong tries** and forgets only **one failure every 10 minutes**. Under that limit, exhausting all 10,000 four-digit combinations takes about **69 days**; a six-digit PIN takes **19 years**.

That entire guarantee rests on one assumption: **the wrong-guess counter cannot be reset**. The TU Berlin team shows that on many x86 machines currently on the market, **that assumption does not hold**:

| **PIN length** | **Under documented lockout** | **Windows Hello attack (AMD)** | **BitLocker attack (Intel)** |
| --- | --- | --- | --- |
| 4 digits | 69 days | **40 hours** | **5 hours**\* |
| 6 digits | 19 years | **168 days** | **20 days** |
| 8 digits | 1,903 years | **46 years** | **6 years** |

*\*BitLocker accepts a four-digit PIN only if an administrator lowers the default six-digit minimum.*

In end-to-end tests on locked, encrypted hardware, the researchers recovered a Windows Hello PIN **after 236 tries in 57 minutes**, and a six-digit BitLocker PIN **after 236 tries in 6 minutes 55 seconds**.

The research documents **two flaws** that break rollback protection, plus a third finding that is arguably more concerning:

1. **Winbond SPI flash leaks its HMAC key** under voltage glitching — allowing **arbitrary counter values to be forged**.
2. **AMD firmware does not verify that a counter increment actually happened** — an interposer board can **suppress the increment unnoticed**. AMD assigned **CVE-2026-43613**.
3. On **AMD machines without RPMC-capable flash**, the researchers **found no rollback protection at all** — writing back an old copy with an ordinary flash programmer resets the counter to zero.

> **Important note on scope:** This is a **physical-access attack** — it requires opening the case, removing the flash chip and wiring in a custom board. **Neither the paper nor the vendor advisories report exploitation outside the lab.** The realistic risk scenario is a **lost, stolen or seized device** — not a remote attack.

---

## Research Context

| **Attribute** | **Detail** |
| --- | --- |
| **Paper title** | Hello Again: Exploiting Replay Protected Monotonic Counters to Brute-Force Windows Hello PINs |
| **Authors** | Christian Werling, Matti Finder, Hans Niklas Jacob, Julian Prein, Niclas Kühnapfel, Jean-Pierre Seifert |
| **Affiliation** | Technische Universität Berlin |
| **Venue** | **ACM CCS '26**; updated arXiv version November 13, 2026 |
| **CVE** | **CVE-2026-43613** (AMD), CVSS 4.0: **5.1** |
| **Released artifacts** | Interposer board firmware, PCB designs, KVM attack scripts, the HMAC keys extracted from the test system's Winbond chip, trace excerpts |
| **Observed in-the-wild exploitation** | **None** |

**On how "ready" the tooling is:** Although the tooling is public, the authors stress that the attacks on RPMC systems are **not plug-and-play**: they require removing the flash chip, soldering in a socket, and using a board of their design. The **AMD-without-RPMC** case is far simpler — it needs only **an off-the-shelf flash programmer**, no custom board.

---

## Technical Analysis

### 1\. Why Can the Lockout Counter Be Reset at All?

#### 1.1. A "firmware" TPM has no secure storage of its own

There are two common kinds of TPM:

| **Type** | **Characteristics** |
| --- | --- |
| **dTPM** (discrete TPM) | A separate chip on the mainboard with **non-volatile (NV) memory inside the chip** |
| **fTPM** (firmware TPM) | Runs on a **coprocessor inside the CPU**: **AMD PSP/ASP**, **Intel CSME** (PTT), or **Microsoft Pluton** |

Most Windows 11 PCs today use an **fTPM**. The problem is that these coprocessors **have no secure on-chip NV memory**. The TPM's entire state — **including the wrong-guess counter** — is **encrypted and integrity-protected** with device-unique keys, then **stored on the mainboard's SPI flash chip** (the same chip that holds the BIOS/UEFI).

Encryption prevents an attacker from **reading or altering** the state. But it **cannot prevent one thing**: **writing back an old copy**.

#### 1.2. Rollback attacks — "snapshot," then "restore"

The attack idea is intuitive:

```plaintext
1. Copy the full flash contents while the wrong-guess counter = 0
2. Guess PINs until close to the lockout threshold (counter rises)
3. Write the old copy back onto the flash chip
   → TPM reloads the old state → counter = 0
4. Repeat from step 2
```

The old copy is **entirely valid** — it is encrypted and signed with the device's own key, because the TPM itself produced it. The TPM has no way to know it is "yesterday's copy" — **unless there is a time source or counter that cannot go backwards**.

#### 1.3. RPMC — a solution designed for exactly this problem

**RPMC (Replay Protected Monotonic Counters)** is a supplementary SPI flash specification from **JEDEC (JESD260, 2021)**, based on a 2013 Intel document. The idea: build into the flash chip **counters that can only increase**, authenticated cryptographically.

| **Component** | **Role** |
| --- | --- |
| **Root key** | Written **exactly once**; authenticates the counter |
| **HMAC session key** | Derived from the root key and "Key Data"; signs each command (HMAC-SHA256) |
| **5 commands** | Write Root Key · Update HMAC Key Register · **Increment Monotonic Counter** · **Request Monotonic Counter** · Read Extended Status |

When the TPM saves new state, it **increments the RPMC counter** and binds the counter value to the state. At boot, the TPM **requests the current counter value** (the Request command — **its response is signed by the chip**) and compares it with the value in the state. If the state is older than the counter → rollback detected.

The Windows 10 and 11 hardware compatibility programs **have required RPMC-style rollback protection** for certain platforms, including **AMD Zen 2 and newer** — though the Windows 11 requirements **do not mandate a specific method**.

> **A detail in the specification itself:** JEDEC states that RPMC **offers no protection against hardware attacks**. "Hello Again" is, in essence, a test of **how wide that self-acknowledged gap is in shipping products**.

---

### 2\. The Interposer Board — Core Component Costs About $5

The attack device is built around a **Raspberry Pi Pico 2W** (RP2350 microcontroller, about **$5**) — using PIO state machines to emulate SPI, **with no expensive FPGA needed**.

| **Component** | **Function** |
| --- | --- |
| Raspberry Pi Pico 2W | Impersonates the flash chip; intercepts and alters commands |
| QSPI-capable multiplexer | Switches the CPU between the real flash and the Pico |
| Logic-level shifter | Converts 3.3 V ↔ 1.8 V |
| MOSFET "crowbar" on the flash supply rail | Produces voltage glitches |
| Logic analyzer | Monitors SPI traffic |

The flash chip is **desoldered from the mainboard** and replaced with a header connecting to the interposer. The team built two versions: **v1** (glitching + interposition) had signal-integrity problems at the Framework laptop's SPI clock speed; **v2** is replay-only with a faster multiplexer, used only for the Framework result. The team also **slowed the flash link in firmware** because the board could not keep up at full speed.

The entire PIN-guessing process was **automated**: a script read the screen, typed PINs and restarted the machine through a **KVM switch**.

---

### 3\. Flaw #1 — Winbond Flash Leaks Its HMAC Key Under Glitching

**Chips involved:**

| **Chip** | **Where** |
| --- | --- |
| Winbond **W25R128JV** (3 V), **W25R128JW** (1.8 V) | Lab testing |
| Winbond **W25R256JW** | Framework Laptop 13 |
| Winbond **74M12JWSIQ** ("Authentication Flash," using RPMC commands) | Lenovo IdeaPad 5 Pro |

**Mechanism:** A short voltage dip (crowbar glitch) while the chip handles a signature causes **the HMAC engine's output to remain in a register readable via the Read Extended Status command**. Two effects were observed:

- After an **Increment**, the chip sometimes returned **a valid signature even though the request was unsigned**.
- After **Update HMAC Key Register**, the chip sometimes returned **the session HMAC key itself** — regardless of whether the update was signed.

**Speed:** with their best settings, the researchers extracted the key in **about one second** — fast enough to glitch **on demand** whenever needed.

**Consequence:** per the authors, this is **almost equivalent to leaking the root key** — the attacker can **forge Request responses with any counter value**. The interposer simply tells the TPM "the counter is still at its old value," with a valid signature.

**Remediation:** requires a **hardware fix** — no register reuse, and never storing secrets in readable registers. Winbond says it is **evaluating a possible silicon revision** but has **committed to no fix or timeline**.

#### An accompanying weakness on the Intel side: constant "Key Data"

On the Intel test laptop, the firmware used a **constant Key Data value of** `0x1` **across every boot and session**. On its own this is a minor weakness. Combined with the Winbond flaw, it means **a single glitch is enough** — the extracted HMAC key **remains usable for every subsequent boot**. Intel PSIRT initially rejected this report as informative, then **reopened it in September 2026** and classified it as **a "security weakness" rather than a "vulnerability."**

---

### 4\. Flaw #2 — AMD Does Not Check That the Counter Actually Incremented (CVE-2026-43613)

This flaw **needs no glitching**, and **works with any RPMC flash**, including chips without Winbond's flaw.

**How the two platforms differ:**

| **Platform** | **After sending an Increment command** |
| --- | --- |
| **Intel** | Sends a **Request** command to **re-read the counter** — the response is signed by the chip |
| **AMD** | **Does not re-read.** Only queries Read Extended Status — which is **unsigned** |

**Exploitation:** the interposer **intercepts the Increment** so the flash chip never receives it, then **answers Read Extended Status with a "success" code**. AMD's firmware accepts this without requiring a signed response. Result: **the counter in the flash chip stays put, while the TPM believes it has increased**.

**AMD's advisory (AMD-SB-7079, November 17, 2026):**

| **Attribute** | **Detail** |
| --- | --- |
| **CVE** | CVE-2026-43613 |
| **CVSS 4.0** | **5.1** — `AV:P/AC:L/AT:N/PR:N/UI:N/VC:N/VI:H/VA:N/SC:N/SI:N/SA:N` |
| **CWE** | CWE-345 — Insufficient Verification of Data Authenticity |
| **Description** | Insufficient authentication of RPMC Increment completion in the ASP firmware could let a physically present attacker suppress counter updates on systems with RPMC-capable SPI flash, bypassing rollback protection |
| **Threat model position** | Attacks requiring physical access are **"outside of the AMD product security threat model"** — but AMD acknowledges **RPMC is intended to stop exactly this type of interposer attack**, and will release firmware updates |
| **Fix timeline** | **November 2026 → February 2027** depending on product line; one "not public" product **has no date** |

**Product lines on the fix list:** Ryzen (desktop, mobile, **Ryzen AI 300/400, AI Max 300/400**), Ryzen Threadripper / Threadripper PRO, Athlon, EPYC 4004/4005, EPYC Embedded, Ryzen Embedded, and the **Z1/Z2** handheld chips. Most consumer fixes target **November–December 2026**; Embedded lines target **February 2027**.

**Two gaps in AMD's advisory worth noting:**

1. The advisory **covers only systems with RPMC flash**. It **says nothing about AMD machines without RPMC** — where the researchers showed rollback works with no custom board (see section 5).
2. The advisory **does not state** whether the fix **blocks downgrading to older, unpatched firmware**. The researchers stress this is a **mandatory condition** — otherwise an attacker simply flashes the old firmware and attacks as before. They propose **eFuses or similar mechanisms** for anti-downgrade protection.

---

### 5\. Finding #3 — AMD Without RPMC: No Rollback Protection Found

On **two AMD desktops without RPMC flash** — an **ASUS TUF B450-Plus (Zen 2)** board and an **MSI PRO B850-P (Zen 5)** board — the researchers only needed to **write back an old SPI image with a flash programmer** to reset the fTPM's lockout counter to zero.

**No interposer. No glitching.** Just an off-the-shelf SPI flash programmer and the ability to open the case.

The authors assess that AMD SoCs **likely have no standalone rollback protection** in the absence of RPMC-style flash — while cautioning that **a sample of two machines may not represent all AMD systems**.

**Compared with Intel:** three Intel systems **without RPMC** (NUC, MSI, Acer) all **detected and rejected** naive rollback. Per the analysis the paper cites, Intel's CSME uses an **anti-replay table** with an **internal timer-based counter** when RPMC is absent.

> **A notable paradox:** On Intel, **having RPMC was worse than not having it**. Intel machines without RPMC rejected rollback; the Intel machine with Winbond RPMC flash was bypassed through the key leak. That is why the authors recommend Intel **revert to its pre-RPMC rollback protection**, at least as a short-term measure.

---

### 6\. Tested Devices

| **Device** | **Platform** | **Flash chip** | **RPMC** | **Naive rollback** |
| --- | --- | --- | --- | --- |
| Intel NUC6i3SYK (2015) | Intel | Winbond W25Q64FV | No | Rejected |
| ASUS TUF B450-Plus (2019) | AMD Zen 2 | Winbond W25Q256JW | No | **Accepted** |
| Lenovo IdeaPad 5 Pro (2021) | AMD Zen 3 | Winbond 74M12JW | Yes (Auth. Flash) | Rejected |
| Lenovo ThinkPad T15 Gen 2 (2021) | Intel | Winbond W25R256JV | Yes | Rejected |
| MSI PRO B760M-P (2024) | Intel | Winbond W25Q256JW | No | Rejected |
| MSI PRO B850-P (2024) | AMD Zen 5 | Macronix MX25U25671G | No | **Accepted** |
| **Framework Laptop 13 (2024)** | Intel | Winbond 25R256JW | Yes | Rejected → **bypassed via leaked HMAC key** |
| **HP OmniBook 3 (2025)** | AMD | GigaDevice 25LR256 | Yes | Rejected → **bypassed via Increment suppression** |
| Acer Aspire 14 AI (2025) | Intel | Macronix MX25U25672G | No | Rejected |

> **How to read this table:** "Rejected" means **naive** rollback (just writing back an old image) failed — **it does not mean the device is immune**. Both the HP OmniBook 3 and the Framework Laptop 13 rejected naive rollback but were bypassed through the two flaws in sections 3 and 4. Moreover, the AMD flaw (section 4) **works with any RPMC flash**, so other AMD machines with RPMC are also in scope until patched.

---

### 7\. End-to-End Tests

#### 7.1. Windows Hello — HP OmniBook 3 (AMD Ryzen AI 5 330), via Increment suppression

- BitLocker-encrypted drive, sitting at the lock screen
- PIN `1235` recovered **after 236 tries in 57 minutes**
- Measured rate: about **4.14 tries per minute**
- Windows forced a restart **every 5 wrong attempts**, but **no lockout delay applied** — because the TPM counter was continually reset

#### 7.2. BitLocker TPM+PIN — Framework Laptop 13 (Intel Core Ultra 5 125H), via Winbond key leak

- PIN `123457` recovered **after 236 tries in 6 minutes 55 seconds**
- Measured rate: about **34.1 tries per minute**
- BitLocker's pre-boot PIN prompt **has no limit of its own** — each counter reset yields **32 fresh tries**

#### 7.3. Full worst-case time table (per the paper)

| **Scenario** | **4 digits** | **6 digits** | **8 digits** |
| --- | --- | --- | --- |
| No rollback, 2-hour recovery\* | 833 days | 228 years | 22,831 years |
| No rollback, 10-minute recovery (Microsoft default) | 69 days | 19 years | 1,903 years |
| **With rollback — Windows Hello** | **40 hours** | **168 days** | **46 years** |
| **With rollback — BitLocker** | **5 hours** | **20 days** | **6 years** |

*\*The researchers observed a 2-hour recovery time on two Intel Core Ultra 5 systems — different from the 10-minute default in Microsoft's documentation.*

> **Two paths independent of the target application:** The authors stress that pairing "AMD ↔ Hello" and "Intel ↔ BitLocker" was only a test choice. **In principle, either method works against either PIN** — stated as an assessment, not cross-tested.

---

### 8\. Microsoft Pluton Is Not Exempt

Microsoft Pluton is Microsoft's own fTPM, running on a dedicated core. The paper assesses that **Pluton on AMD is affected by the Increment suppression flaw** just as AMD's own fTPM is — because **running Microsoft code on a dedicated core does not change the fact that state is still stored on external flash through the same controller**. Note: this is **the authors' assessment**, **not directly demonstrated** by testing.

---

### 9\. What Does an Attacker Gain From a PIN?

| **Target** | **Consequence** |
| --- | --- |
| **BitLocker PIN (TPM+PIN)** | **Decrypts the entire drive** — the data-protection layer for a lost or stolen device |
| **Windows Hello PIN** | Unlocks **device-bound cryptographic keys** that Hello protects |
| **Windows Hello for Business** | Access to Hello-based **enterprise SSO** |
| **Some configurations** | May **satisfy an authentication step otherwise protected by MFA** |

That last point is why this goes beyond "losing a laptop": in an enterprise using Hello for Business, **a stolen laptop can become an entry point into the organization's identity system**.

---

### 10\. Secondary Flash Chip Findings

| **Finding** | **Vendor** | **Severity per the authors** |
| --- | --- | --- |
| **Integer overflow** — the 32-bit counter wraps from maximum to 0 (about 4–6 weeks in practice) | Winbond, Macronix, GigaDevice | **Minor** for NV state rollback, but **serious** for counter-based firmware anti-downgrade |
| **Counter corruption** — glitching during Request returns a value lower than the real one yet **still validly signed** (~10% success) | GigaDevice | **Low** |

The RPMC specification **does not define behavior when a counter reaches its maximum** — unlike eMMC's RPMB, which rejects further increments. The researchers sent a proposal to JEDEC in **April 2025**; as of publication, **the specification has not been updated**.

---

## Risk Summary

| **Risk Dimension** | **Level** | **Rationale** |
| --- | --- | --- |
| **Exploitation prerequisites** | **High (difficult)** | Requires **physical access**, opening the case, removing the flash chip; RPMC systems need a custom board and soldering skill |
| **Observed in-the-wild exploitation** | **None** | Neither the paper nor vendor advisories report exploitation outside the lab |
| **Degradation of protection** | **Very High** | Six-digit BitLocker PIN: **19 years → 20 days**; four-digit Hello PIN: **69 days → 40 hours** |
| **Impact if exploited** | **Critical** | Full drive decryption; access to device-bound keys, enterprise SSO, possibly bypassing MFA |
| **Hardware scope** | **Broad** | AMD machines using fTPM (most modern Ryzen lines), Intel machines with Winbond RPMC flash, very likely Pluton on AMD |
| **Patch status** | **Partial** | AMD: timeline **Nov 2026–Feb 2027**; **Winbond: no commitment**; **Intel: not deemed an Intel vulnerability**; Microsoft: only raised the recommended PIN length |
| **Gap no one has claimed** | **Concerning** | **AMD machines without RPMC** — not covered by AMD's advisory; the attack needs only an ordinary flash programmer |
| **Tooling availability** | **Medium–High** | Firmware, PCB designs and scripts are public; the core component costs about $5 |

---

## IOCs & Artifacts

*This is hardware vulnerability research, not a malware campaign —* ***there are no network IOCs or malware hashes***\*. Instead, this section lists exposure indicators and signs of physical tampering.\*

### Vulnerability identifiers

```plaintext
CVE-2026-43613   AMD — RPMC Increment Suppression (CVSS 4.0: 5.1, CWE-345)
AMD-SB-7079      AMD security bulletin (November 17, 2026)
Intel 2026-11-16-001   Intel security announcement "Hello Again" (November 16, 2026)
```

### Exposure check on AMD machines (from Linux)

Linux kernel **5.19 and later** exposes RPMC status through the `ccp` driver. Boot a live Linux environment and run:

```bash
grep . /sys/bus/pci/drivers/ccp/*/rpmc_*
```

- The output shows **whether RPMC is present and enabled**.
- **Missing files do not prove that RPMC is absent.**
- The `fwupd` tool also reports this status as **"AMD SPI Replay protections."**

**Interpreting the result:**

| **Result** | **Meaning** |
| --- | --- |
| RPMC present and enabled | Affected by **CVE-2026-43613** until patched firmware is installed |
| RPMC absent / disabled | Per the paper's results, there may be **no rollback protection at all** — **higher** risk, and **AMD's advisory does not cover** this case |

### Exposure check on Intel machines

**There is no software check.** Intel offers no equivalent interface. The only way is to **open the machine and look for a Winbond flash chip marked** `25R` **or** `74M`, or log SPI traffic.

### Flash chips confirmed in the research

```plaintext
Winbond W25R128JV / W25R128JW / W25R256JV / W25R256JW   (25R series — RPMC)
Winbond 74M12JW / 74M12JWSIQ                             ("Authentication Flash")
GigaDevice 25LR256                                       (HP OmniBook 3 — counter corruption flaw)
```

### Signs of physical tampering to check

- **Tamper seals or case screws** showing signs of removal
- The **SPI flash chip** showing traces of **resoldering**, or replaced with a **socket/header**
- **Unfamiliar wires** attached near the BIOS chip area
- **Abnormally frequent restarts** in the event log while no user was present
- **SPI flash link speed** in firmware settings lowered from its default

---

## MITRE ATT&CK Mapping

*Note: ATT&CK is designed around network intrusion; physical hardware attacks map only partially. The table below is for reference.*

| **Tactic** | **ID** | **Technique** | **Manifestation in the research** |
| --- | --- | --- | --- |
| **Initial Access** | T1200 | Hardware Additions | Inserting an interposer board between the CPU and the SPI flash chip |
| **Defense Evasion** | T1542.001 | Pre-OS Boot: System Firmware | Overwriting SPI flash contents with an old image (rollback) |
| **Defense Evasion** | T1562 | Impair Defenses | Neutralizing the TPM's dictionary-attack lockout |
| **Credential Access** | T1110.001 | Brute Force: Password Guessing | Automated Windows Hello / BitLocker PIN guessing via KVM |
| **Credential Access** | T1552.004 | Unsecured Credentials: Private Keys | Extracting the HMAC key from the Winbond chip via voltage glitching |
| **Collection** | T1005 | Data from Local System | Decrypting the BitLocker drive after recovering the PIN |
| **Credential Access / Lateral Movement** | T1550 | Use Alternate Authentication Material | Using Hello device-bound keys for enterprise SSO (authors' assessment, not tested) |

---

## Expert Assessment

**The core lesson: a short PIN is only safe when guesses are limited, and that limit is only as strong as where the counter lives.** Microsoft allows four-digit PINs on sound reasoning: 10,000 combinations is far too few with no limit, but enough if the TPM allows 32 tries and then makes you wait. This research breaks no cryptography and no TPM — it simply shows that **on most machines today, that counter sits on a flash chip outside the CPU, and can be copied back**. A security guarantee built on "the counter cannot go backwards" collapses completely when that assumption is false.

**The RPMC paradox on Intel deserves emphasis.** RPMC was introduced to **strengthen** rollback protection. Yet on the Intel machines tested, **machines without RPMC rejected rollback**, while the machine with Winbond RPMC flash was bypassed. A new security component, implemented with a flaw, can **replace** an older mechanism that was working well — and leave the system weaker. That is why the authors' recommendation to Intel sounds counterintuitive but is entirely sensible: **go back to the pre-RPMC mechanism**.

**The responsibility gap is the most concerning part.** Look again at each party's response:

- **AMD:** fixes the Increment suppression flaw, but treats physical attacks as "outside the threat model," and the advisory **does not address machines without RPMC**.
- **Intel:** "the research did not demonstrate a vulnerability in an Intel component" — because the Winbond flaw was required.
- **Winbond:** evaluating, **no committed fix or timeline**.
- **Microsoft:** rated "moderate," no CVE, no bounty; committed to **raising the recommended minimum PIN to six digits**.
- **JEDEC:** no specification update after more than a year.

Taken individually, each response is defensible. Taken together, **the end user is left with a system in which every party believes the fault lies with another**. The vulnerability lives **at the intersection of components from multiple vendors** — JEDEC's specification, Winbond's chip, AMD/Intel firmware, Microsoft's PIN policy — and precisely because of that, no one owns the whole problem. This is a textbook hardware supply-chain vulnerability, and such vulnerabilities tend to **persist for a very long time**.

**On "requires physical access" — do not underestimate it.** The argument "you have to open the machine, so the risk is low" misses one thing: **BitLocker TPM+PIN exists precisely to defend against someone with physical access**. If the attacker cannot touch the machine, BitLocker has almost nothing to do. In other words, the scenario in this research **is exactly the scenario BitLocker was designed for** — and in that scenario, the default six-digit PIN now holds for roughly **20 days**. For a deliberately stolen device, 20 days is not an obstacle.

**A detail worth acknowledging about the authors' care:** the paper clearly separates what was **tested** (two attack paths on two laptops, two AMD desktops with naive rollback) from what is **only assessed** (Pluton, cross-pairing methods with applications, possible MFA bypass). We have preserved that boundary throughout this article.

**For Vietnam, the caveat comes first: the research has no country-level data, and there is no recorded in-the-wild exploitation anywhere.** Even so, three aspects of the local context make these results worth attention from organizations:

**First — AMD laptops are very common in the Vietnamese market**, particularly in mainstream and office segments, including the new Ryzen AI lines. This is exactly the group of devices on AMD's fix list — and the fix depends on **each laptop maker shipping an updated BIOS**, a step that is often slow and rarely installed proactively by users.

**Second — government agencies and businesses are deploying BitLocker to meet data-protection requirements**, many with the default six-digit PIN or TPM-only mode. For laptops issued to officials and staff who travel frequently, the scenario of **a device lost or seized for an extended period** is real. This group should consider **lengthening PINs** now, without waiting for patches.

**Third — the lifecycle of retired devices.** Laptops and workstations being decommissioned, resold or sent for warranty repair often **still have encrypted drives that have not been securely wiped**. BitLocker encryption used to be considered sufficient reassurance. With these results, any device that **has left the organization's control** should be treated as at risk, and decommissioning procedures should **securely erase data** rather than rely on encryption alone.

---

## Recommendations

### Urgent — actionable now, no patch required

1. **Lengthen PINs by at least two digits.** This is the most effective measure and **depends on no vendor**: each added digit multiplies attack time **tenfold**.
   - **Windows Hello:** from 4 to **at least 6** (40 hours → 168 days)
   - **BitLocker TPM+PIN:** from 6 to **at least 8** (20 days → 6 years)
2. **Windows Hello for Business:** set the **Minimum PIN length** policy under `Computer Configuration > Administrative Templates > System > PIN Complexity`. If unset, Windows requires six digits by default.
3. **BitLocker:** raise the **Configure minimum PIN length for startup** policy (allowed range 4 to 20 digits; default 6). Consider enabling **Allow enhanced PINs for startup** to permit letters and symbols — **expanding the keyspace by many orders of magnitude** compared with digits alone.
4. **If you use BitLocker in TPM-only mode (no PIN):** move to **TPM+PIN** with a long PIN. Note that TPM+PIN relies on the same lockout this research undermines, so **PIN length is the decisive factor**.

### Short term — monitor and update firmware

1. **AMD machines:** install the firmware/BIOS update from your PC maker **as soon as it is available for your model**. AMD's timeline: **November 2026–February 2027**. AMD's fix reaches users through **each OEM** (HP, Lenovo, Dell, ASUS…) — check your maker's support page proactively.
2. **Check RPMC status** on AMD machines with `grep . /sys/bus/pci/drivers/ccp/*/rpmc_*` from a live Linux environment, or via `fwupd`. Machines **without RPMC** should get priority physical protection because they are **outside the scope of AMD's advisory**.
3. **Intel machines with Winbond RPMC flash:** **no fix has been committed**. Intel advises **contacting your system manufacturer** about available updates.
4. **In the meantime:** per AMD's guidance — **keep devices under physical control** and **watch for signs of physical tampering**.

### Long term — physical control and device lifecycle

1. **Apply tamper-evident seals** to laptop case screws for devices holding sensitive data, and **inspect them periodically**, especially after a device has left your control (repairs, travel, seizure).
2. **Lost-device procedure:** treat every lost device as **data that may be decrypted within weeks**. Immediately **revoke keys and sessions tied to the device** in your identity system (Entra ID / Active Directory), and **disable the device** for Hello for Business and SSO.
3. **Decommissioning:** **securely erase** drives (crypto-erase or overwrite) rather than relying on BitLocker encryption alone.
4. **When procuring new devices:** for high-security needs, consider devices with a **discrete TPM (dTPM)** — the lockout counter lives in memory inside the TPM chip, **independent of external flash**. Note that dTPMs carry their own risks (e.g., TPM bus sniffing), so evaluate against your organization's threat model.
5. **For servers and critical workstations:** where firmware allows, enable **firmware anti-downgrade protection** so patches cannot be undone by reflashing an older BIOS.

---

## References

- [Hello Again: Exploiting Replay Protected Monotonic Counters to Brute-Force Windows Hello PINs — Werling, Finder, Jacob, Prein, Kühnapfel, Seifert (TU Berlin), arXiv 2611.18126 (November 13, 2026)](https://arxiv.org/abs/2611.18126v1)
- [ACM CCS '26 conference version — DOI 10.1145/3830454.3846631](https://doi.org/10.1145/3830454.3846631)
- [rpmc_attacks — the research team's firmware, board designs and scripts (GitHub)](https://github.com/mattifinder/rpmc_attacks)
- [AMD-SB-7079: Replay Protected Monotonic Counters (RPMC) Increment Suppression Vulnerability — AMD (November 17, 2026)](https://www.amd.com/en/resources/product-security/bulletin/amd-sb-7079.html)
- [Intel Security Announcement 2026-11-16-001: "Hello Again" — Intel (November 16, 2026)](https://www.intel.com/content/www/us/en/security-center/announcement/intel-security-announcement-2026-11-16-001.html)
- [TPM fundamentals (default lockout parameters) — Microsoft Learn](https://learn.microsoft.com/en-us/windows/security/hardware-security/tpm/tpm-fundamentals)

---
