← Back to Feed CACHED · 2026-07-01 06:48:07 · CACHE_KEY CVE-2026-20461
CVE-2026-20461 · CWE-787 · Disclosed 2026-07-01

In Modem

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

A malformed cellular packet from a fake tower knocks the modem off the network — nothing more

CVE-2026-20461 is an out-of-bounds write (CWE-787) in the MediaTek modem firmware caused by a missing bounds check in a packet parser. An attacker who stands up a rogue base station the target device attaches to can craft a message that overruns a modem-side buffer. MediaTek describes the impact as remote denial of service, with no user interaction and no elevated privileges required. The affected list is huge — 33 SoCs including MT6895/6897/6899, MT6989/6991/6993, MT8676/8678, MT8798, MT8893 and more — covering flagship 5G smartphones, tablets, Chromebooks, and connected-car/IoT modules shipped from ~2022 onward. Patch IDs are MOLY01267281 / MOLY01318201, delivered to OEMs at least 60 days before disclosure and rolling into the July 2026 Android Security Bulletin dependent-patch cycle.

MediaTek labels this HIGH. That rating reflects the modem's ring-0-equivalent position on the baseband, not the actual demonstrated impact. The advisory is explicit: the outcome is DoS, not RCE, not code execution in the AP kernel, not SIM/IMSI theft. Baseband OOB writes *can* escalate to code execution, but MediaTek does not claim that here and no public PoC demonstrates it. Weighed against the operational reality — attacker must stand up an SDR + fake gNodeB within RF range, victims lose service until the modem resets — this is a MEDIUM. It becomes interesting only in targeted-surveillance scenarios where a rogue BTS is already the operating model.

"Baseband DoS via rogue base station across dozens of MediaTek SoCs — noisy, proximity-bound, and DoS-only. MEDIUM, not High."
02 · The Attack Path

4 steps from start to impact.

STEP 01

Stand up a rogue base station in RF range of the target

Attacker deploys an SDR-driven fake gNodeB/eNodeB (srsRAN, Open5GS, YateBTS, or commercial IMSI-catcher gear like the Stingray/Hailstorm family). The rogue cell must broadcast a stronger signal than legitimate carriers on a band the target device supports so the modem selects it during cell reselection.
Conditions required:
  • Physical proximity (typically <500m)
  • SDR hardware (USRP B210/X310 or similar, ~$1.5k-$25k)
  • Target band knowledge (n41, n78, LTE B2/B4/B12 etc.)
Where this breaks in practice:
  • Broadcasting on licensed spectrum is a federal crime in most jurisdictions
  • Modern 5G SA enforces mutual authentication — pure downgrade attacks are increasingly detectable
  • Requires physical presence — does not scale beyond targeted operations
Detection/coverage: No enterprise EDR sees this. Carrier-side rogue-BTS detection (SnoopSnitch, CellularPrivacy) is consumer-grade and rarely deployed on fleet devices.
STEP 02

Force target device to attach to the rogue cell

Standard IMSI-catcher playbook: broadcast a spoofed carrier PLMN/TAC with high signal strength, optionally jam legitimate bands to trigger reselection. The MediaTek modem in idle-mode reselection will camp on the strongest allowed cell that presents a valid SIB1.
Conditions required:
  • Victim device in idle or actively scanning
  • SIM allows roaming onto spoofed PLMN, or attacker mimics home PLMN
Where this breaks in practice:
  • 5G SA with SUCI concealment and mutual auth limits full attach
  • Corporate MDM profiles restricting roaming reduce success rate
  • iOS and modern Android surface 'connected to unsecured network' hints in some builds
Detection/coverage: Modem crash logs and rapid signal-drop patterns can be surfaced by MTD (Zimperium, Lookout) — most fleets don't run MTD.
STEP 03

Deliver malformed NAS/RRC message that trips the missing bounds check

Attacker sends a crafted layer-3 message (likely a NAS or RRC IE with an oversized length field or nested structure) that the modem parser copies without validating destination buffer size. The exact vulnerable parser has not been publicly disclosed; MediaTek internal reporting only cites CWE-787 in the modem stack.
Conditions required:
  • Device has completed enough of the attach to accept the vulnerable message type
  • Attacker has reverse-engineered the parser or has a working PoC (none public as of July 2026)
Where this breaks in practice:
  • Requires baseband reverse engineering skill — small pool of practitioners (Comsecuris, TaszkLab, Grant Hernandez, Amat Cama tier)
  • MediaTek modem firmware is stripped/obfuscated and varies per SoC — one PoC does not port cleanly across all 33 chipsets
Detection/coverage: Invisible to host OS. Only baseband forensic capture (rare in enterprise) would catch it.
STEP 04

Modem crashes → cellular DoS

The OOB write corrupts adjacent modem heap/stack state. Per MediaTek's own writeup the demonstrated outcome is denial of service: modem resets, loses cellular data and voice, sometimes takes the whole handset down until reboot. Escalation to code execution on the baseband is theoretically possible for a CWE-787 primitive but is not claimed by the vendor and has no public PoC.
Conditions required:
  • Successful delivery of the malformed message
Where this breaks in practice:
  • DoS ends when victim moves out of RF range or reboots — impact is transient
  • No pivot to AP-side (application processor / Android userspace) documented
Detection/coverage: Users notice service loss. No structured telemetry back to IT unless MDM tracks radio state.
03 · Intelligence Metadata

The supporting signals.

In-the-wild exploitationNone observed. Not on CISA KEV. No public reports of active use as of 2026-07-01.
Public PoCNone published. MediaTek reports the finding as internally discoveredno external researcher credit, no GitHub repo, no conference talk.
EPSSNot yet scored (CVE published today, 2026-07-01). Baseband OOB CVEs historically settle around EPSS <0.5%.
KEV statusNot listed. CISA has never KEV-listed a MediaTek baseband DoS CVE.
CVSSNo authoritative CVSS score published. MediaTek severity label: High. NVD entry pending. Vector inferable as roughly AV:A/AC:H/PR:N/UI:N/S:U/C:N/I:N/A:H (adjacent-radio, high complexity, availability-only).
Affected chipsets33 SoCs: MT2737, MT6813, MT6835, MT6858, MT6878, MT6879, MT6886, MT6895, MT6896, MT6897, MT6899, MT6980, MT6983, MT6985, MT6988, MT6989, MT6990, MT6991, MT6993, MT8673, MT8676, MT8678, MT8755, MT8775, MT8792, MT8793, MT8795T, MT8796, MT8798, MT8863, MT8873, MT8883, MT8893 — Dimensity 8000/9000 series, Kompanio, Genio.
Fixed versionModem patch IDs MOLY01267281 and MOLY01318201. Delivered to OEMs ≥60 days before 2026-07-01. Rollout depends on OEM (Xiaomi, Oppo, Vivo, Motorola, Samsung-A, Google Chromebook OEMs) shipping the July 2026 vendor patch level.
Exposure telemetryShodan/Censys/GreyNoise are irrelevant — attack surface is RF, not IP. No internet-facing indicator. Exposure is proximity-based.
Disclosure2026-07-01 MediaTek July 2026 Product Security Bulletin, source: Internal.
ReporterMediaTek internal security team. No external credit.
04 · The Call

noisgate verdict.

Final Verdict
DOWNGRADED to MEDIUM (5.3/10)

The single most decisive factor is impact ceiling: MediaTek only claims denial-of-service, and there is no public evidence of RCE escalation on the baseband. Combined with the mandatory rogue base station prerequisite — physically-present attacker, illegal RF broadcast, deep baseband skill — the realistic risk to a 10,000-host fleet sits firmly in MEDIUM territory, not HIGH.

HIGH on impact classification (DoS only, per vendor)
HIGH on absence of public PoC and in-the-wild use
MEDIUM on future escalation potential to baseband RCE
LOW on exact vulnerable message parser (not disclosed)

Why this verdict

  • Impact ceiling is DoS: vendor advisory is explicit — no confidentiality, no integrity, no persistence, no AP-side compromise. A modem crash equals lost signal until reboot.
  • Attack vector is RF-adjacent, not remote-internet: requires a rogue gNodeB within RF range. This eliminates opportunistic mass exploitation and confines threat to targeted operations (nation-state, journalist/dissident targeting).
  • No public PoC, no in-the-wild use, not KEV: internal MediaTek find, disclosed today, zero exploitation telemetry, EPSS effectively baseline.
  • Role multiplier — smartphone/tablet endpoint: The chain succeeds in the *typical* role (employee handset, tablet, Chromebook), blast radius is loss of cell service on one device. No lateral movement, no fleet effect.
  • Role multiplier — connected vehicle / IoT modem (MT2737, MT8755, MT8863 telematics): If exploited against a vehicle TCU during operation, DoS on cellular could disable emergency call / OTA / fleet-management uplink. Blast radius stays scoped to that vehicle; no supply-chain or fleet pivot. Not a floor-raising role because population is small and impact is still availability-only.
  • Role multiplier — no canonical high-value role: MediaTek modems do not host IdP, PKI, hypervisor, backup, or kernel-mode security agent functions. No HIGH/CRITICAL floor triggers.
  • Friction stack: illegal RF broadcast + physical presence + baseband RE + 33-SoC parser variance. Every layer prices out commodity attackers.

Why not higher?

HIGH would require either a plausible path to baseband code execution (MediaTek does not claim it, no PoC demonstrates it) or a widely-exploitable network vector (this needs a rogue base station, not the public internet). Vendor's own HIGH label conflates the modem's privileged position with actual demonstrated impact — for enterprise defenders managing 10,000 hosts, transient cellular DoS on a subset of Android/Chromebook devices does not warrant HIGH triage.

Why not lower?

LOW would undersell the fact that the affected SoC list is enormous — Dimensity 9000/8000-class flagships ship in tens of millions of devices — and the attack requires zero user interaction. Targeted-surveillance actors with rogue-BTS capability (already-established TTPs: Stingray, Hailstorm, criminal IMSI-catcher gray market) can weaponize DoS as a coercion tool. MEDIUM is right; below MEDIUM ignores the non-zero targeted risk.

05 · Compensating Control

What to do — in priority order.

  1. Ship the July 2026 OEM patch level via MDM policy — Intune / Jamf / Workspace ONE / Kandji: set minimum Android security patch level to 2026-07-01 (or the earliest OEM level that includes MOLY01267281 / MOLY01318201) with a 365-day noise-free remediation window per the noisgate MEDIUM SLA. Non-compliant devices should lose access to sensitive apps via Conditional Access.
  2. Enforce 5G SA-only on high-risk personas — For execs, legal, M&A, and threat-model-relevant staff, push a carrier profile / APN config that disables 2G and prefers 5G SA. This blocks the classic downgrade-to-legacy attach path that many IMSI-catchers rely on. On Android 14+ use the Cellular Security settings toggle 'Require 5G'.
  3. Deploy Mobile Threat Defense (Zimperium zIPS, Lookout MES, Wandera) to high-risk fleets — MTD agents flag suspicious cell behavior (rapid signal changes, unknown PLMN attach, baseband reset patterns). Not perfect but the only signal enterprises get for rogue-BTS activity. Deploy alongside patch rollout within the same 365-day MEDIUM window.
  4. Add rogue-BTS awareness to executive protection briefings — For travel to high-surveillance jurisdictions, brief on Wi-Fi-first mode, avoiding unknown cell hand-offs, using eSIM burners. Especially relevant if you also have unpatched Qualcomm/MediaTek baseband issues stacking.
What doesn't work
  • Corporate WAF / IPS / NGFW — completely useless. Attack lives on the RF layer, never touches your IP network.
  • VPN always-on — irrelevant. The bug triggers before any IP traffic; the modem is DoS'd out from under the VPN.
  • Blocking specific IPs / domains via DNS — no C2 infrastructure involved; there is nothing to block.
  • Faraday cases — technically stop the attack but also stop the phone from being a phone. Only viable for stored devices, not in-use fleet.
  • Google Play Protect / MDM app scanning — the vulnerability is in modem firmware, not an app. App-layer scanners have zero visibility.
06 · Verification

Crowdsourced verification payload.

Run on an Intune-connected auditor workstation with Microsoft.Graph.DeviceManagement module and Global Reader / Intune Reader rights. Invoke as .\Check-CVE-2026-20461.ps1 -MinPatchLevel '2026-07-01'. This inventories Android/Chromebook devices, flags those with MediaTek SoCs below the July 2026 patch level, and outputs VULNERABLE / PATCHED / UNKNOWN per device. Requires no credentials on the endpoint itself — everything reads through Graph.

noisgate-verify.ps1
POWERSHELLREAD-ONLYSAFE
#Requires -Modules Microsoft.Graph.DeviceManagement
<#
.SYNOPSIS
  Audit Intune-managed Android/Chromebook devices for CVE-2026-20461
  (MediaTek modem OOB write, patched in July 2026 OEM patch level).
.PARAMETER MinPatchLevel
  Minimum acceptable Android security patch level, format YYYY-MM-DD.
#>
param(
    [Parameter(Mandatory=$true)][string]$MinPatchLevel
)

$ErrorActionPreference = 'Stop'

# Affected MediaTek SoC families (substrings that appear in device hardware fields)
$AffectedSoCs = @(
    'MT2737','MT6813','MT6835','MT6858','MT6878','MT6879','MT6886',
    'MT6895','MT6896','MT6897','MT6899','MT6980','MT6983','MT6985',
    'MT6988','MT6989','MT6990','MT6991','MT6993','MT8673','MT8676',
    'MT8678','MT8755','MT8775','MT8792','MT8793','MT8795T','MT8796',
    'MT8798','MT8863','MT8873','MT8883','MT8893',
    'Dimensity','Kompanio','Genio'
)

try {
    Connect-MgGraph -Scopes 'DeviceManagementManagedDevices.Read.All' -NoWelcome
} catch {
    Write-Error "Graph auth failed: $_"
    exit 2
}

$minDate = [datetime]::ParseExact($MinPatchLevel,'yyyy-MM-dd',$null)
$devices = Get-MgDeviceManagementManagedDevice -All -Filter "operatingSystem eq 'Android' or operatingSystem eq 'ChromeOS'"

$results = foreach ($d in $devices) {
    $soc = "$($d.Manufacturer) $($d.Model) $($d.HardwareInformation.Product)"
    $isAffectedSoC = $AffectedSoCs | Where-Object { $soc -match $_ }
    $status = 'UNKNOWN'

    if (-not $isAffectedSoC) {
        $status = 'PATCHED'  # not an affected SoC family
    } else {
        $patchStr = $d.AndroidSecurityPatchLevel
        if ($patchStr) {
            try {
                $patchDate = [datetime]::ParseExact($patchStr,'yyyy-MM-dd',$null)
                if ($patchDate -ge $minDate) { $status = 'PATCHED' }
                else { $status = 'VULNERABLE' }
            } catch { $status = 'UNKNOWN' }
        }
    }

    [pscustomobject]@{
        DeviceName    = $d.DeviceName
        User          = $d.UserPrincipalName
        OS            = $d.OperatingSystem
        SoCHint       = $soc.Trim()
        PatchLevel    = $d.AndroidSecurityPatchLevel
        Status        = $status
    }
}

$results | Sort-Object Status,DeviceName | Format-Table -AutoSize
$vulnCount = ($results | Where-Object Status -eq 'VULNERABLE').Count
Write-Host ""
Write-Host "VULNERABLE: $vulnCount" -ForegroundColor Red
Write-Host "PATCHED   : $(($results | Where-Object Status -eq 'PATCHED').Count)" -ForegroundColor Green
Write-Host "UNKNOWN   : $(($results | Where-Object Status -eq 'UNKNOWN').Count)" -ForegroundColor Yellow

if ($vulnCount -gt 0) { exit 1 } else { exit 0 }
07 · Bottom Line

If you remember one thing.

TL;DR
Monday morning: file this under MEDIUM backlog, not a fire drill. Per the noisgate mitigation SLA, MEDIUM carries no mitigation deadline — go straight to the noisgate remediation SLA of ≤365 days: push the July 2026 Android/ChromeOS OEM patch level to every MediaTek-based device in your fleet through Intune/Jamf/Workspace ONE compliance policy over the next 12 months, prioritizing exec/legal/M&A personas in the first 30 days. Do not build custom detections, do not open a P1 bridge, do not chase rogue-BTS threat models unless you already run a mobile threat intel program. If credible intel emerges of an actual RCE PoC on the baseband (watch Comsecuris, TaszkLab, and Google Project Zero blogs), re-triage to HIGH and collapse the deadline; until then, the biggest risk to your fleet this quarter is somewhere else.

Sources

  1. MediaTek July 2026 Product Security Bulletin
  2. OffSeq Threat Radar — CVE-2026-20461
  3. CWE-787: Out-of-bounds Write
  4. Android Security Bulletin — July 2026
  5. CISA Known Exploited Vulnerabilities Catalog
  6. FIRST EPSS Model
  7. srsRAN / rogue base station research context
  8. Comsecuris baseband research (background reading on modem RCE feasibility)
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.