A kernel parser that trusts whatever an app hands it, dressed up in a remote-RCE costume
CVE-2026-43724 is one of the kernel-side input-validation bugs Apple closed in the iOS 26.5.2 / iPadOS 26.5.2 cycle on June 29, 2026. Per Apple's release notes, the affected kernel path could be coerced by a malicious application into unexpected termination, kernel-memory write, kernel-state disclosure, or memory corruption — i.e., a sandbox-escape / EOP primitive, not a network listener bug. Affected versions are all iOS/iPadOS builds prior to 26.5.2; the fix is improved input sanitization on the affected syscall/IOKit interface.
The vendor severity (CRITICAL 9.8 with AV:N/AC:L/PR:N/UI:N) does not match the technical reality. A kernel bug reachable only from an *installed application* is AV:L/PR:L, not AV:N/PR:N. Apple's CNA frequently publishes inflated vectors on kernel CVEs in batch advisories; the realistic CVSS sits in the 7.0–7.8 band. The capability is still serious — kernel write means full device compromise, MDM bypass, and Lockdown Mode is *not* a guaranteed mitigation — but the threat model is malicious-app-installed, not drive-by-network.
4 steps from start to impact.
Deliver a malicious app to the target device
- Attacker-controlled native code executing in an iOS app sandbox
- Device on iOS/iPadOS < 26.5.2
- User or MDM accepts/installs the app
- App Store review + notarization filters obviously malicious binaries
- Enterprise/developer signing requires a stolen cert or compromised provisioning profile
- Lockdown Mode blocks unverified configuration profiles and tightens app behavior
- MDM-managed corporate fleets typically allow-list apps
Invoke the vulnerable kernel interface with malformed input
- Knowledge of the specific vulnerable interface (currently undisclosed by Apple)
- Ability to call the interface from sandbox (no entitlement gate)
- Apple has not published the affected component name — diffing 26.5.2 IPSW is required
- If the interface is entitlement-gated, sandboxed App Store apps cannot reach it without a stolen entitlement
sysdiagnose panic logs containing the affected kext name post-exploitation; EDR-for-mobile (Lookout, Zimperium, Jamf Protect) flags anomalous kernel panics.Corrupt kernel memory / leak kernel state
- Reliable primitive (write > read > panic for offensive value)
- Heap-grooming control sufficient to land the corruption on a useful object
- PAC, PPL, and SPTM on A12+/M1+ raise the cost of converting a single primitive into kernel R/W
- KASLR + page-protection layer require an additional infoleak
Escape sandbox, persist, exfiltrate
- Kernel R/W achieved in step 3
- Knowledge of post-exploit chain for the target iOS build
- Reboot evicts most non-persistent implants
- Lockdown Mode restricts persistence surfaces (configuration profiles, MDM, message attachments)
- Apple Threat Notifications can flag known mercenary-spyware infrastructure
The supporting signals.
| In-the-wild status | No public exploitation reported as of 2026-07-01. Apple did *not* flag this CVE with the "may have been actively exploited" language used for the chain-exploited kernel bugs in the same advisory. |
|---|---|
| Proof-of-concept | No public PoC. Weaponization requires binary-diffing iOS 26.5.1 vs 26.5.2 IPSWs; expect researcher write-ups from @_simo36, @i41nbeer, or Google P0 within 30-60 days. |
| EPSS | 0.0018 (≈ 0.18%) — bottom-quartile likelihood of exploitation in the next 30 days. |
| CISA KEV | Not listed. No federal exploitation evidence. |
| CVSS vector (vendor) | CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H = 9.8. Disputed — true vector is closer to AV:L/AC:L/PR:L/UI:N/S:C/C:H/I:H/A:H (~7.8), as the bug is reachable only from an installed app. |
| Affected versions | All iOS/iPadOS builds prior to 26.5.2, across iPhone XS and later, all iPad Pro models, iPad Air 3+, iPad 7+, iPad mini 5+. |
| Fixed version | iOS 26.5.2 / iPadOS 26.5.2, released 2026-06-29. No backports — Apple does not branch-patch older majors for non-actively-exploited kernel bugs. |
| Scanning / exposure data | Not internet-discoverable — no Shodan/Censys/GreyNoise signal possible for a local kernel bug. Enterprise exposure measured by MDM compliance dashboards (Intune, Jamf, Workspace ONE). |
| Disclosure | 2026-06-29 via Apple Security Releases HT-26.5.2. |
| Reporter | Not publicly credited in the initial advisory (Apple commonly back-fills credit in the next bulletin). |
noisgate verdict.
Downgraded from CRITICAL because the single most decisive factor is attack-vector mislabeling: this is a local app-delivered kernel bug, not the network-reachable unauth-RCE the 9.8 vector implies. The floor stays at HIGH because the primitive is kernel-memory corruption on a mobile fleet — a successful chain ends in full-device takeover, Keychain theft, and MDM-token exfiltration.
Why this verdict
- Attack vector is local, not network. Apple's own description ("an application may be able to…") is
AV:L/PR:L, which alone drops the score by ~2 points from the published 9.8. - No active exploitation and no KEV listing. Apple flags exploited kernel bugs explicitly in the same bulletin; CVE-2026-43724 was not flagged.
- EPSS 0.0018 + no public PoC means the realistic 30-day exploitation likelihood is near zero outside of mercenary-spyware shops who would burn this on high-value targets, not enterprise mass.
- Role multiplier — executive / high-risk mobile users: chain succeeds, ends in device takeover and Keychain/MDM-token theft. Blast radius is per-device, not per-fleet — high-value-user exposure keeps the floor at HIGH but does not justify CRITICAL.
- Role multiplier — corporate fleet workstation-class iPhone: chain succeeds but requires a malicious app reaching the device, which MDM allow-listing materially blocks. Blast radius stays at the single device.
Why not higher?
Not CRITICAL because (a) the attack requires an installed malicious application — not a network packet, not a malicious URL, not an email — which gates the realistic threat actor pool to those who can ship apps to a target (mercenary spyware, supply-chain compromise of a legitimate app, or social-engineering a sideload); and (b) there is no exploitation evidence, no KEV, no public PoC. The vendor's 9.8 reflects worst-case capability, not realistic exposure.
Why not lower?
Not MEDIUM because the primitive — kernel-memory write / kernel-state disclosure — is the exact capability that mercenary-spyware vendors weaponize, and the affected fleet (every non-26.5.2 iPhone/iPad in the enterprise) is large. On VIP and high-risk-user devices the blast radius is unacceptable to defer beyond the standard 30-day HIGH window.
What to do — in priority order.
- Push iOS/iPadOS 26.5.2 via MDM compliance policy now — Set Intune/Jamf/Workspace ONE minimum-OS to
26.5.2, block corporate resource access (Conditional Access / Jamf Connect) for devices below that build. Per the noisgate mitigation SLA for HIGH, complete within 30 days; for VIP/executive devices, treat as 3 days. - Enable Lockdown Mode on VIP, journalist, executive, and high-risk-traveler devices — Lockdown Mode shrinks the kernel-reachable attack surface (restricts configuration profiles, MDM enrollment changes, message attachment processing) and is the single most effective control against the mercenary-spyware threat actor most likely to weaponize this CVE.
- Tighten app allow-listing and disable sideloading / enterprise-cert profiles — In Intune/Jamf, restrict installation to the corporate VPP catalog; revoke unused enterprise developer profiles. The attack requires app delivery — choke the delivery channel.
- Deploy mobile EDR (Jamf Protect, Lookout, Zimperium, Wandera) to the fleet — These tools collect kernel panic logs and detect known commercial-spyware indicators; they will not block this bug but will catch post-exploit behavior and surface jailbreak-state anomalies.
- Run iVerify / MVT scans on VIP devices monthly — MVT (Mobile Verification Toolkit) and iVerify hunt for Pegasus/Predator-class IOCs and detect anomalous kernel/process state consistent with kernel-corruption exploitation.
- WAF / network IPS — the vuln is local-app-to-kernel; nothing traverses the corporate network in an exploitable form.
- Email gateway scanning — there is no email-borne payload; the delivery channel is the App Store / TestFlight / enterprise signing.
- Disabling Safari / WebKit — this is a kernel bug, not a WebKit bug; browser hardening is unrelated.
- MFA — irrelevant; the attacker is exploiting kernel-input handling, not authenticating to anything.
Crowdsourced verification payload.
Run on your MDM admin workstation (macOS or Linux) against an exported device inventory CSV. Requires read access to the Intune/Jamf device list and jq. Example invocation: ./check-43724.sh devices.csv where devices.csv has columns device_id,os_version.
#!/usr/bin/env bash
# noisgate verification: CVE-2026-43724 (iOS/iPadOS kernel input-validation)
# Fixed in iOS/iPadOS 26.5.2. Anything below is VULNERABLE.
set -euo pipefail
FIXED_MAJOR=26
FIXED_MINOR=5
FIXED_PATCH=2
if [ $# -ne 1 ]; then
echo "usage: $0 <mdm-export.csv> (columns: device_id,os_version)" >&2
exit 2
fi
CSV="$1"
[ -r "$CSV" ] || { echo "cannot read $CSV" >&2; exit 2; }
vuln=0; patched=0; unknown=0
while IFS=, read -r dev ver; do
[ "$dev" = "device_id" ] && continue
ver_clean="${ver//\"/}"
if [[ ! "$ver_clean" =~ ^([0-9]+)\.([0-9]+)(\.([0-9]+))?$ ]]; then
echo "$dev UNKNOWN ($ver_clean)"
unknown=$((unknown+1)); continue
fi
M=${BASH_REMATCH[1]}; m=${BASH_REMATCH[2]}; p=${BASH_REMATCH[4]:-0}
if (( M < FIXED_MAJOR )) || \
(( M == FIXED_MAJOR && m < FIXED_MINOR )) || \
(( M == FIXED_MAJOR && m == FIXED_MINOR && p < FIXED_PATCH )); then
echo "$dev VULNERABLE ($ver_clean)"
vuln=$((vuln+1))
else
echo "$dev PATCHED ($ver_clean)"
patched=$((patched+1))
fi
done < "$CSV"
echo "---"
echo "VULNERABLE=$vuln PATCHED=$patched UNKNOWN=$unknown"
if (( vuln > 0 )); then exit 1; fi
if (( unknown > 0 )); then exit 3; fi
exit 0
If you remember one thing.
26.5.2, and stage a forced update with a Conditional Access block on corporate resources for non-compliant devices. Per the noisgate mitigation SLA for HIGH, complete the compliance push (or app allow-list tightening + Lockdown Mode on VIPs as a stopgap) within 30 days; per the noisgate remediation SLA, every managed device must be on 26.5.2 or later within 180 days. For executives, board members, journalists in your fleet, and frequent high-risk travelers, treat this as CRITICAL-class on those devices and enable Lockdown Mode plus push the patch within 3 days. Do *not* burn cycles on WAF rules or email-gateway tuning — the delivery channel is the app, not the network.Sources
What defenders are saying.
Crowdsourced verification outputs.
Results submitted by users who ran the verification payload against their environment.