← Back to Feed CACHED · 2026-05-17 09:42:19 · cache_key CVE-2025-29912
CVE-2026-25262 · CWE-123 · Disclosed 2026-04-20

Qualcomm BootROM Sahara/EDL write-what-where secure boot bypass

ASSESSED — NOISGATE V0.5
Vendor
Reassessed
Verdict:
01 · The Real Story

This is a master key hidden behind a locked service door, not a brick through your internet-facing window

CVE-2026-25262 is a BootROM CWE-123 write-what-where flaw in Qualcomm's Sahara handling inside Emergency Download Mode (EDL). Public Kaspersky research says it affects all versions of the MDM9x07, MDM9x45, MDM9x65, MSM8909, MSM8916, MSM8952, and SDX50 chipset series, letting an attacker with physical access bypass the secure boot chain and run code with maximum privilege before the OS comes up.

Reality check: the *technical* impact is ugly, but the *operational* attack path is narrow. There is no useful vendor baseline to compare against, and the only public CVSS-like vector we found collapses to 0.0 because CVSS v3.1 treats AV:P awkwardly; that badly understates the per-device impact, but it still does not make this a Monday-morning fleetwide fire drill because the exploit requires hands-on device custody and an EDL/USB servicing path.

"ASSESSED AT MEDIUM: devastating on one device, but physical custody and USB/EDL access keep this out of the emergency patch lane."
02 · The Attack Path

4 steps from start to impact.

STEP 01

Get physical custody of the device

The attacker first needs the target device in hand long enough to attach USB or otherwise place it into a servicing/recovery workflow. Kaspersky explicitly frames realistic scenarios as third-party repair, customs inspection, theft-and-return, insider handling, or supply-chain staging rather than remote internet attack.
Conditions required:
  • Attacker can physically handle the target device
  • Device uses an affected Qualcomm chipset
  • Attacker has enough uninterrupted time for cabling and recovery-mode interaction
Where this breaks in practice:
  • This is not remotely reachable over IP
  • Custody events are auditable in many enterprises
  • Many corporate endpoints never leave controlled hands or are tamper-evident
Detection/coverage: Traditional vuln scanners will miss this step; best signals are chain-of-custody logs, depot/repair records, USB forensic artifacts, and asset inventory that maps devices to affected chipsets.
STEP 02

Enter EDL and speak Sahara with standard tooling

Once the device is in Qualcomm EDL, the attacker can use existing ecosystem tooling such as QFIL/QPST, qdl, or public projects like bkerler/edl to communicate over Sahara. This is important: there is no public turnkey exploit repo for *this* CVE in our search, but the transport and servicing workflow are already well understood and publicly documented.
Conditions required:
  • Device can be placed into EDL or is already in a recoverable servicing state
  • Attacker has USB access and host tooling
  • Any OEM controls around service access are bypassed or absent
Where this breaks in practice:
  • Some deployments lock down service workflows or require authorized depots
  • Gaining EDL can vary by product design and board exposure
  • Field devices may be sealed or physically hardened
Detection/coverage: Look for devices enumerating as Qualcomm EDL/HS-USB recovery hardware during maintenance; EDR is usually blind because this occurs before the normal OS security stack is active.
STEP 03

Trigger BootROM write-what-where and break the chain of trust

Kaspersky's advisory describes a write-what-where condition in BootROM chunk validation that can let a physical attacker bypass secure boot and execute arbitrary code with maximum privileges. This is the decisive impact step: compromise happens below the OS, so standard kernel hardening, MDM policy, and userland security controls come too late.
Conditions required:
  • Exploit works against the specific chipset implementation
  • Attacker can deliver crafted data over Sahara/EDL
  • Target is one of the named affected series or a similarly affected Qualcomm derivative
Where this breaks in practice:
  • Attack complexity is rated high by the public advisory
  • Exploit development still requires deep chipset knowledge
  • Cross-device reliability may differ by OEM integration
Detection/coverage: There is little to no reliable scanner coverage here; post-compromise validation is mostly indirect via anomalous boot behavior, forensic comparison, or trusted reflashing and power-cycle testing.
STEP 04

Plant pre-OS code or backdoor the application processor

After secure boot is bypassed, the attacker can access sensitive data and in some cases fully compromise the device, including backdooring higher layers on the application processor. Kaspersky also warns that a normal reboot may not be trustworthy if malware simulates restart behavior, making incident response messy once compromise is suspected.
Conditions required:
  • Successful low-level code execution from step 3
  • Target stores high-value credentials, files, sensors, or tokens
  • Attacker wants persistence, covert collection, or later-stage pivoting
Where this breaks in practice:
  • Blast radius is usually one device at a time
  • A full power loss may evict some volatile implants
  • Operational value depends on what that specific device can access
Detection/coverage: Behavioral clues may include unexplained heat, battery drain, traffic anomalies, or sensor misuse, but confidence is weak; pre-boot implants are a blind spot for standard EDR/AV.
03 · Intelligence Metadata

The supporting signals.

In-the-wild statusNo authoritative active-exploitation signal found. CISA KEV does not list this CVE, and the sources reviewed describe attack scenarios and impact rather than confirmed campaigns.
Proof-of-concept availabilityNo public turnkey CVE-specific exploit repo found in this search. However, public research exists via Kaspersky's advisory/blog/Black Hat Asia talk, and generic Qualcomm EDL/Sahara tooling like bkerler/edl lowers the implementation barrier.
EPSSNo public EPSS score located during this review. That is plausible for a newly disclosed hardware CVE with limited enrichment and no broad exploitation telemetry yet.
KEV statusNot in CISA KEV as of this assessment; no dateAdded found in the public catalog.
Published CVSS / vectorKaspersky ICS CERT publishes CVSS:3.1/AV:P/AC:H/PR:N/UI:N/S:U/C:H/I:H/A:H with score 0.0. Treat that as a CVSS artifact, not a sane enterprise priority signal: the impact is severe, but AV:P drags the numeric score into nonsense territory.
Affected versionsAll versions of Qualcomm MDM9x07, MDM9x45, MDM9x65, MSM8909, MSM8916, MSM8952, and SDX50 per Kaspersky ICS CERT. Kaspersky also notes other Qualcomm-based chips may be affected, but only these series are explicitly confirmed.
Fixed versions / patch statusNo field patch for existing chips. Kaspersky states BootROM is immutable on already-manufactured devices and says Qualcomm plans to remove the flaw in future chip revisions instead.
Exposure / scanning realityNo meaningful Shodan/Censys-style internet surface. This is a USB/physical EDL path, not a normal network service, so external attack-surface tools are the wrong lens; your exposure lives in device inventory, repair-chain handling, and field custody.
Disclosure timelineKaspersky ICS CERT advisory published 2026-04-20; Kaspersky press release followed 2026-04-23; Qualcomm was reportedly notified in March 2025 and acknowledged in April 2025.
Researcher / reporting orgReported by Alexander Kozlov and Sergey Anufrienko of Kaspersky ICS CERT; public research presented at Black Hat Asia 2026.
04 · The Call

noisgate verdict.

Final Verdict
= UNCHANGED to MEDIUM (5.9/10)

The single biggest downward pressure is the attacker position requirement: this is a physical-custody, USB/EDL servicing-path exploit, not an unauthenticated remote bug. We keep it at MEDIUM rather than LOW because a successful hit lands below the OS, bypasses secure boot, and can create a very high-impact compromise on sensitive mobile, IoT, automotive, or industrial endpoints.

HIGH Physical-access and EDL prerequisites materially reduce reachable population
MEDIUM Per-device impact is severe once exploitation succeeds
LOW Public evidence of broad in-the-wild exploitation

Why this verdict

  • Physical custody is the gate. The attacker needs hands-on access to the device and a servicing/recovery interaction path, which immediately removes this from the common internet-exposed exploit queue.
  • This usually implies a prior failure in process or trust. Internal handling, third-party repair, customs seizure, depot workflows, or supply-chain access are all narrower and less scalable than network exploitation, so the reachable enterprise population drops hard.
  • The impact is too deep to ignore. Secure-boot bypass and pre-OS arbitrary code execution with maximum privilege create a high-value compromise on each affected device, especially where credentials, telemetry, microphones/cameras, payment flows, or operational control functions are present.

Why not higher?

It is not higher because the exploit path is not remotely reachable, appears to require physical custody, and carries high technical complexity. We also found no KEV listing and no strong evidence of broad in-the-wild campaigns, which matters when you are triaging thousands of assets.

Why not lower?

It is not lower because the vulnerability sits in BootROM, below the OS trust boundary, and can bypass secure boot with maximum privilege on the device. The lack of a real patch for existing silicon is another amplifier: if an affected asset matters, the residual risk is real even without mass exploitability.

05 · Compensating Control

What to do — in priority order.

  1. Lock the repair chain — Use only authorized depots, maintain documented chain-of-custody, and forbid unsupervised third-party repair for affected assets. For a MEDIUM verdict there is no mitigation SLA; implement this during normal process hardening, but do it before the next repair, redeployment, or travel cycle for any affected fleet.
  2. Inventory affected chipsets — Map devices, modems, tablets, IoT nodes, automotive units, and industrial endpoints to the named Qualcomm series so you know where this risk actually exists. There is no mitigation SLA for MEDIUM — go straight into normal asset-governance work and finish inventory well inside the 365-day remediation window.
  3. Restrict EDL and depot access — Physically secure service cables, flashing stations, and any workflow that can place devices into EDL or connect to them in recovery mode. For MEDIUM there is no mitigation SLA; fold this into hardware-security baselines and maintenance SOPs before the next field-service event.
  4. Force full power loss after suspicious custody events — If an affected device has been out of custody, repaired, or handled suspiciously, perform a trusted inspection and ensure a complete power loss rather than trusting a normal reboot alone. There is no mitigation SLA for MEDIUM, but this should happen immediately at the point of return-to-service because the control is event-driven, not calendar-driven.
What doesn't work
  • Routine OS patching does not fix immutable BootROM code on already-shipped chips.
  • Network segmentation alone does not address a USB/physical pre-boot attack path.
  • A warm reboot is not a reliable cleanup step; Kaspersky explicitly warns that only complete power loss can guarantee a clean restart in some scenarios.
  • EDR/AV alone is not enough because compromise happens before the normal OS telemetry and protection stack is active.
06 · Verification

Crowdsourced verification payload.

Run this on the target device itself if it exposes a POSIX shell, or through ADB shell for Android-based assets. Example: adb push cve-2026-25262-check.sh /data/local/tmp/ && adb shell sh /data/local/tmp/cve-2026-25262-check.sh; no root is required for the common checks, but elevated privileges may improve access to device-tree paths.

noisgate-verify.sh
BASHREAD-ONLYSAFE
#!/bin/sh
# CVE-2026-25262 chipset exposure check
# Purpose: identify whether the current device appears to use one of the
# publicly named affected Qualcomm chipset families.
# Output: VULNERABLE / PATCHED / UNKNOWN
# Exit codes: 1=VULNERABLE, 0=PATCHED, 2=UNKNOWN

set -u

TMP=""
cleanup() {
  [ -n "$TMP" ] && [ -f "$TMP" ] && rm -f "$TMP"
}
trap cleanup EXIT INT TERM

TMP="/tmp/cve-2026-25262.$$"
: > "$TMP" 2>/dev/null || TMP="./cve-2026-25262.$$"
: > "$TMP" || {
  echo "UNKNOWN: cannot create temp file"
  exit 2
}

append_file() {
  f="$1"
  if [ -r "$f" ]; then
    # Device-tree files may contain NUL bytes.
    tr '\000' ' ' < "$f" >> "$TMP" 2>/dev/null
    printf '\n' >> "$TMP"
  fi
}

append_cmd() {
  # shellcheck disable=SC2068
  if command -v "$1" >/dev/null 2>&1; then
    "$@" >> "$TMP" 2>/dev/null || true
    printf '\n' >> "$TMP"
  fi
}

# Common Linux / Android identification sources
append_file /proc/device-tree/model
append_file /proc/device-tree/compatible
append_file /sys/firmware/devicetree/base/model
append_file /sys/firmware/devicetree/base/compatible
append_file /sys/devices/soc0/family
append_file /sys/devices/soc0/machine
append_file /sys/devices/soc0/soc_id
append_file /sys/devices/soc0/revision
append_cmd uname -a
append_cmd cat /proc/cpuinfo
append_cmd getprop ro.board.platform
append_cmd getprop ro.boot.hardware
append_cmd getprop ro.boot.hardware.sku
append_cmd getprop ro.boot.qcomsoc_id
append_cmd getprop ro.hardware
append_cmd getprop ro.product.board
append_cmd getprop ro.product.device
append_cmd getprop ro.soc.manufacturer
append_cmd getprop ro.soc.model
append_cmd lspci -nn
append_cmd lsusb
append_cmd dmesg

DATA=$(tr '[:upper:]' '[:lower:]' < "$TMP")

if [ -z "$DATA" ]; then
  echo "UNKNOWN: no platform-identification data collected"
  exit 2
fi

# Exact/publicly named affected families from Kaspersky ICS CERT.
# Include common derivative naming seen in products where helpful.
PATTERNS="
mdm9x07
mdm9207
mdm9607
mdm9x45
mdm9645
mdm9x65
mdm9655
msm8909
msm8916
msm8952
sdx50
snapdragon x50
"

for p in $PATTERNS; do
  echo "$DATA" | grep -Eq "(^|[^a-z0-9])${p}([^a-z0-9]|$)" && {
    echo "VULNERABLE: matched affected Qualcomm chipset identifier '${p}'"
    exit 1
  }
done

# If it is clearly Qualcomm but not one of the named affected families,
# keep the result conservative because public sources note other Qualcomm-based
# chips may also be affected.
echo "$DATA" | grep -Eq 'qualcomm|qcom|snapdragon|msm|mdm|sdx' && {
  echo "UNKNOWN: Qualcomm platform detected, but no exact match to the publicly confirmed affected families"
  exit 2
}

echo "PATCHED: no evidence this host uses one of the publicly confirmed affected Qualcomm chipset families"
exit 0
07 · Bottom Line

If you remember one thing.

TL;DR
Monday morning, do not treat this like a remote edge-service emergency; treat it like a hardware trust and custody problem. For a MEDIUM verdict there is no noisgate mitigation SLA — go straight to the 365-day remediation window for inventory, procurement, and replacement planning, but immediately tighten repair/depot/field-service handling for any affected fleet because there is no real patch for shipped silicon. Use the noisgate remediation SLA as the planning clock for identifying affected devices and moving future purchases to non-vulnerable chip revisions, while event-driven controls like custody checks and trusted return-to-service inspections happen whenever a device leaves your hands.

Sources

  1. Kaspersky ICS CERT advisory KLCERT-25-012
  2. Kaspersky ICS CERT vulnerabilities index
  3. Kaspersky press release
  4. Kaspersky technical blog
  5. Qualcomm May 2026 security bulletin landing page
  6. Black Hat Asia 2026 presentation
  7. Public Qualcomm EDL/Sahara tooling
  8. CISA Known Exploited Vulnerabilities Catalog
Peer Review

What defenders are saying.

Submit a review attribution: handle + country only
0 flags selected · stored anonymously
Validation Results

Crowdsourced verification outputs.

Results submitted by users who ran the verification payload against their environment.