← Back to Feed CACHED · 2026-07-01 17:02:48 · CACHE_KEY CVE-2026-20458
CVE-2026-20458 · CWE-787 · Disclosed 2026-07-01

In Modem

ASSESSED — NOISGATE V0.5
Vendor
Reassessed
Verdict:
Do you agree?
01 · The Real Story

A bug that lets the cell tower punch a hole in your phone's modem — but only if the attacker owns the tower

CVE-2026-20458 is a classic CWE-787 out-of-bounds write in MediaTek's Modem (MOLY) firmware, triggered when the modem parses a malformed message from the serving cell. The flaw affects 53 chipsets spanning low-end feature-phone SoCs (MT6739) up through flagship Dimensity parts (MT6989, MT6991, MT8798) — essentially MediaTek's entire fielded lineup. MediaTek fixed the issue with internal patch MOLY01402160 (Issue MSV-7298) and shipped it to OEMs at least two months before the July 1, 2026 public disclosure.

MediaTek scored this HIGH (7.5, CVSS:3.1/AV:A/AC:H/PR:N/UI:N/S:U/C:H/I:H/A:H). That score is *directionally correct but overstated for enterprise fleet risk* — the AV:A adjacent-network vector is the honest part (attacker must be in RF range with a rogue base station), and AC:H reflects the real engineering effort of standing up a malicious eNB/gNB. For a fleet manager, the addressable population is not the datacenter — it's the corporate-issued handset and BYOD program. That reshapes the risk profile substantially.

"Baseband OOB write on 53 MediaTek chipsets — but exploitation demands a rogue base station in RF range. Fleet risk is BYOD, not server room."
02 · The Attack Path

4 steps from start to impact.

STEP 01

Stand up a rogue base station within RF range

Attacker deploys a rogue LTE/5G-NR eNB or gNB (typically a srsRAN or OpenAirInterface stack running on a USRP B210/X310 SDR) configured to advertise the target carrier's PLMN with a stronger signal than the legitimate cell. Modern Stingray-class commercial IMSI catchers achieve the same result. The target UE must be physically within a few hundred meters at most.
Conditions required:
  • Attacker owns SDR + amplifier + directional antenna
  • Attacker knows target's carrier PLMN/TAC
  • Target device is in RF range and not locked to 5G-SA only
Where this breaks in practice:
  • Regulated spectrum — jamming/rogue-eNB deployment is a felony in most jurisdictions
  • Physical proximity requirement means one-target-at-a-time, not fleet-scale
  • Modern handsets increasingly warn on downgrade to unauthenticated 2G/3G
Detection/coverage: Enterprise MDMs (Intune, Jamf, MobileIron) do not detect rogue base stations. Specialized apps (SnoopSnitch, CellularPrivacy) can flag anomalies but require root on Android.
STEP 02

Coerce UE to camp on the rogue cell

The rogue cell broadcasts SIB1/MIB with attractive parameters; the target UE performs RRC connection and NAS attach. If the attacker cannot complete authentication (they don't have SIM Ki), they still get to send crafted L2/L3 signaling messages before the auth challenge fails — which is where the pre-auth modem parser sits.
Conditions required:
  • UE RRC state machine accepts the rogue cell
  • Rogue cell can send the vulnerable message pre-authentication
Where this breaks in practice:
  • 5G-SA with SUCI concealment reduces the pre-auth attack surface
  • Carrier-locked auth timers close the window quickly
Detection/coverage: Baseband logs are not exposed to userland; forensic recovery requires vendor tools.
STEP 03

Deliver malformed modem message to trigger OOB write

Attacker sends the crafted signaling message containing the length/offset that fails the missing bounds check. The MOLY modem firmware performs an out-of-bounds write into the modem's heap, corrupting adjacent structures. Weaponization requires reverse engineering of the specific MOLY build for the target chipset — non-trivial but well within reach of state-grade research teams (Project Zero, Unit 42, NCC Group have published comparable baseband work).
Conditions required:
  • Attacker has reversed target MOLY build to identify controllable write primitive
  • Modem heap layout is predictable enough for reliable corruption
Where this breaks in practice:
  • MOLY builds vary per OEM/region — one exploit does not cover all 53 chipsets
  • MediaTek has added baseband hardening (stack cookies, W^X) in recent MOLY revisions
Detection/coverage: No practical detection — modem crashes may surface as radio panics in dmesg if pr_err from mtk_modem driver reaches kernel log.
STEP 04

Achieve code execution in the modem, pivot to AP

Successful primitive yields code execution in the baseband processor (isolated from the Application Processor by the SMEM/RPMSG boundary). To reach the Android/user data, attacker chains a second baseband-to-AP escape — historically demonstrated via shared memory (SMEM) corruption or IPC parser bugs on the Linux side. Without that second stage, impact is limited to modem-level compromise (call intercept, SMS intercept, location tracking, DoS).
Conditions required:
  • Second-stage baseband-to-AP escape available
  • Kernel is not running full lockdown / hypervisor-enforced isolation
Where this breaks in practice:
  • Modern Android (14+) with pKVM strongly isolates baseband from AP
  • Apple/iOS not affected — MediaTek modems are Android-only
Detection/coverage: MTD (Mobile Threat Defense) tools like Zimperium, Lookout, Wandera can flag anomalous modem behavior post-compromise but not the exploit itself.
03 · Intelligence Metadata

The supporting signals.

In-the-wild statusNone observed. No public exploit code, no known campaigns as of 2026-07-02.
PoC availabilityNo public PoC. Baseband reverse-engineering research from Google Project Zero and NCC Group has demonstrated the general class.
KEV statusNot KEV-listed.
EPSSNot yet scored by FIRST (disclosed 2026-07-01, model needs ~72h).
CVSS vectorCVSS:3.1/AV:A/AC:H/PR:N/UI:N/S:U/C:H/I:H/A:H7.5 HIGH. AV:A = adjacent network (RF range only); AC:H = requires attacker-controlled base station.
CWECWE-787 Out-of-bounds Write.
Affected chipsets53 MediaTek SoCs including MT6739, MT676x/677x/678x/679x series, Dimensity flagships (MT6989, MT6991), and MT8xxx tablet parts. See MediaTek July 2026 PSB.
FixMediaTek internal patch MOLY01402160 (Issue MSV-7298), shipped to OEMs ≥2 months before 2026-07-01 disclosure. Downstream availability depends on OEM/carrier (Samsung, Xiaomi, Oppo, Vivo, Motorola, Nokia HMD).
Disclosure2026-07-01 via MediaTek Product Security Bulletin (July 2026). Reporter: External (unnamed).
Exposure populationEnterprise-fleet relevance is confined to Android handsets on MediaTek silicon — roughly ~30% of global Android shipments, heavily concentrated in mid-tier and emerging-market devices. Apple, Qualcomm-powered Samsung/Pixel flagships not affected.
04 · The Call

noisgate verdict.

Final Verdict
DOWNGRADED to MEDIUM (5.8/10)

The single most decisive factor is the AV:A adjacent-network prerequisite — exploitation requires a rogue base station in physical RF range of a specific target, which collapses the attack from fleet-scale to one-target-at-a-time and puts it firmly in the surveillance-actor category rather than commodity ransomware. The affected component is a handset modem, not an identity, hypervisor, backup, or kernel-mode security role, so there is no fleet-multiplier that would drag the floor up to HIGH.

HIGH Vulnerability class and affected chipset list (from MediaTek PSB)
MEDIUM Real-world exploitability — depends on unpublished MOLY build details
LOW Downstream OEM patch availability across 53 chipsets and dozens of vendors

Why this verdict

  • Adjacent-network only: AV:A with a rogue-BTS prerequisite eliminates internet-scale exploitation. Every successful attack costs the adversary physical proximity — this is not going to be commodity malware.
  • High attack complexity is real: AC:H reflects the genuine engineering lift — SDR + MOLY reverse-engineering per chipset variant. Cost floor sits at the state-actor / commercial-surveillance-vendor tier (NSO, Cytrox, Candiru class).
  • Role multiplier — handset baseband: Blast radius on successful chain is *one device* → call/SMS/location intercept, potentially AP escape on unhardened builds. It is NOT identity, hypervisor, PAM, backup, CA, EDR, or edge appliance. No fleet floor triggers — MEDIUM ceiling is honest.
  • Modem hardening trend: MediaTek MOLY has added stack cookies and W^X across recent releases, and Android 14+ pKVM isolates baseband from AP. Full chain (baseband RCE → AP takeover) is not free even after this primitive.
  • Patch pipeline latency is the actual pain: OEM/carrier patching for baseband firmware historically runs 3–9 months behind MediaTek's OEM push. For an enterprise MDM fleet, tracking OEM patch level per device is the real work, not the exploit.

Why not higher?

HIGH would require either fleet-scale reach or a high-value role. AV:A + AC:H + rogue-BTS operational cost forecloses fleet scale; a handset modem is a personal-device role, not identity/edge/hypervisor. No KEV, no public PoC, no campaign attribution — nothing to justify a HIGH floor.

Why not lower?

LOW would ignore that the primitive is a pre-auth heap corruption on the modem with no user interaction (UI:N) and full CIA impact on the modem. This is exactly the class of bug used in past commercial-surveillance chains (Amnesty/Citizen Lab Predator/Pegasus disclosures), so for executive-protection or high-risk-user populations the risk is non-trivial. MEDIUM correctly reflects that.

05 · Compensating Control

What to do — in priority order.

  1. Enforce OEM security patch level via MDM conditional access — In Intune/Jamf/Workspace ONE, set a conditional-access policy requiring Android security patch level ≥ 2026-07-05 (the July Android security bulletin merge point) for corporate resource access. This is MEDIUM verdict, so no mitigation SLA under the noisgate SLA — but staging the conditional-access rule within 60 days aligns with the 365-day remediation window and forces OEM/carrier patch pull-through.
  2. Disable 2G on managed handsets — Android 14+ exposes a disable 2G toggle via DevicePolicyManager.setCellularTwoGDisabled. Push this via MDM. Most rogue-BTS attacks downgrade to 2G to bypass mutual auth — killing 2G raises the attacker's engineering bar and cost meaningfully.
  3. Executive/high-risk-user program: rotate to lockdown-mode-capable devices — For board members, M&A staff, security research, and legal — devices with iOS Lockdown Mode or Samsung Knox equivalent, or non-MediaTek silicon, materially reduce baseband attack surface. Complete within 90 days for named high-risk-user list.
  4. Inventory MediaTek-powered corporate handsets — Pull chipset field from your MDM (Intune ships hardware.SoC; Jamf exposes Hardware.Model). Cross-reference against the 53-chipset MediaTek list. You cannot prioritize what you cannot see.
  5. Update Mobile Threat Defense signatures — Verify Lookout, Zimperium, or Wandera has telemetry for baseband anomalies and rogue-BTS heuristics. Coordinate with vendor for CVE-2026-20458-specific detection updates within 30 days.
What doesn't work
  • Corporate VPN or ZTNA — the attack is on the radio stack, below IP. VPN never sees the traffic.
  • Application-layer WAF, EDR on the handset, or MDM compliance policies — none inspect modem firmware behavior.
  • SIM PIN / eSIM provisioning controls — the vulnerable parser runs before SIM auth, so PIN doesn't gate it.
  • MFA — irrelevant; the compromise is on the device, so the attacker inherits the authenticated session.
06 · Verification

Crowdsourced verification payload.

Run this on a target Android handset via adb shell (USB debugging enabled, developer mode on). Invocation: adb shell sh /sdcard/check_cve_2026_20458.sh. No root required — reads only public getprop values. Reports VULNERABLE if the SoC is on the MediaTek list AND the Android security patch level is older than 2026-07-05.

noisgate-verify.sh
BASHREAD-ONLYSAFE
#!/system/bin/sh
# noisgate verification: CVE-2026-20458 (MediaTek MOLY OOB write)
# Requires: adb shell access, no root
# Output: VULNERABLE / PATCHED / UNKNOWN

set -u

AFFECTED_SOCS="MT2716 MT2737 MT6739 MT6761 MT6762 MT6763 MT6765 MT6767 MT6768 MT6769 MT6771 MT6779 MT6781 MT6783 MT6785 MT6789 MT6813 MT6835 MT6858 MT6878 MT6879 MT6881 MT6886 MT6895 MT6896 MT6897 MT6899 MT6980 MT6982 MT6983 MT6985 MT6986 MT6988 MT6989 MT6990 MT6991 MT6993 MT8666 MT8667 MT8668 MT8673 MT8676 MT8678 MT8755 MT8765 MT8766 MT8768 MT8775 MT8781 MT8786 MT8788 MT8789 MT8792 MT8793 MT8796 MT8798 MT8863 MT8873 MT8883 MT8893"

FIX_PATCH_DATE="2026-07-05"

SOC=$(getprop ro.hardware 2>/dev/null; getprop ro.board.platform 2>/dev/null; getprop ro.soc.model 2>/dev/null)
PATCH=$(getprop ro.build.version.security_patch 2>/dev/null)
MANUFACTURER=$(getprop ro.product.manufacturer 2>/dev/null)
MODEL=$(getprop ro.product.model 2>/dev/null)

if [ -z "$SOC" ] || [ -z "$PATCH" ]; then
  echo "UNKNOWN: could not read SoC or patch level"
  exit 2
fi

echo "Manufacturer: $MANUFACTURER"
echo "Model:        $MODEL"
echo "SoC:          $SOC"
echo "Patch level:  $PATCH"

is_affected=0
for s in $AFFECTED_SOCS; do
  echo "$SOC" | grep -qi "$s" && is_affected=1 && break
done

if [ "$is_affected" -eq 0 ]; then
  echo "PATCHED: SoC not on MediaTek CVE-2026-20458 affected list"
  exit 0
fi

# Compare YYYY-MM-DD lexicographically
if [ "$PATCH" \< "$FIX_PATCH_DATE" ]; then
  echo "VULNERABLE: MediaTek SoC detected, patch level $PATCH < $FIX_PATCH_DATE"
  exit 1
else
  echo "PATCHED: MediaTek SoC detected, patch level $PATCH >= $FIX_PATCH_DATE"
  exit 0
fi
07 · Bottom Line

If you remember one thing.

TL;DR
Monday morning: pull your MDM inventory, filter to MediaTek-powered handsets against the 53-chipset list, and identify the delta between current OEM patch level and 2026-07-05. This is a MEDIUM verdict — under the noisgate mitigation SLA there is no mitigation deadline (go straight to the 365-day remediation window), but you should stage the two cheap wins in parallel: push the *Disable 2G* MDM policy fleet-wide within 30 days, and set a conditional-access policy requiring the July 2026 Android patch level within 60 days. Under the noisgate remediation SLA you have until 2027-07-02 to reach OEM-patched builds across the fleet — but expect the tail on budget Android OEMs (feature-phone MT6739 devices, no-name tablets) to slip past that window; plan the retirement/replacement of unpatchable devices now. For executive-protection and high-risk-user populations, accelerate: rotate off MediaTek silicon or to Lockdown-Mode-capable devices within 90 days.

Sources

  1. MediaTek Product Security Bulletin — July 2026
  2. OffSeq Threat Radar — CVE-2026-20458
  3. MITRE CWE-787: Out-of-bounds Write
  4. Google Project Zero — Over the Air baseband research
  5. Android — Disable 2G enterprise control
  6. NCC Group — Baseband exploitation research
  7. srsRAN Project — open-source RAN stack (attack tooling reference)
  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.