The AI concierge forgot to check guest badges, so strangers could read the VIP registry
CVE-2026-86987 is an authorization flaw in Apple Intelligence, the on-device and Private Cloud Compute AI subsystem that powers features like Writing Tools, notification summaries, Genmoji, and Siri enhancements. Insufficient entitlement validation allowed a remote attacker to access sensitive user data — the kind of data Apple Intelligence processes: emails, messages, photos, calendar events, and notification content. The bug affects all Apple Intelligence-capable devices running iOS or iPadOS prior to 27.2, meaning iPhone 15 Pro and later, iPad Pro 12.9-inch 5th gen+, iPad Pro 11-inch 2nd gen+, iPad Air 5th gen+, and iPad mini 7th gen. The fix shipped on October 26, 2026 in iOS/iPadOS 27.2 as part of a massive 163-CVE security update.
Apple has not assigned a CVSS score, and NVD analysis is still pending. VulDB tentatively scored it 4.3, but the remote attacker / sensitive user data pairing warrants closer scrutiny. The description's hedging language ('may be able to') plus the narrow device population (only Apple Intelligence-capable hardware) and purely confidentiality-scoped impact keep this firmly in MEDIUM territory. There is no evidence of in-the-wild exploitation, no public proof-of-concept, and the EPSS sits at the 8.7th percentile (0.19%). This is a 'patch in your normal iOS cycle' issue, not an emergency — but the data at stake on executive devices means it shouldn't collect dust either.
3 steps from start to impact.
Identify Apple Intelligence-capable target
- Target device runs iOS/iPadOS < 27.2
- Apple Intelligence is enabled on the device
- Network path to the target exists
- Apple Intelligence requires iPhone 15 Pro+ or specific iPads — excludes ~60% of the active iPhone fleet as of late 2026
- User must have opted in to Apple Intelligence during setup; not all eligible users enable it
- No Shodan/Censys/FOFA fingerprint exists for individual iOS devices
Craft malicious request bypassing entitlement validation
- Knowledge of the Apple Intelligence API surface or IPC interface
- Ability to deliver crafted payloads to the target device
- No public PoC or technical writeup exists — attack methodology is not documented
- Apple's sandboxing and code-signing generally prevent arbitrary IPC from untrusted processes
- Exploitation likely requires chaining with another vector (malicious web content, a compromised app, or network-level interaction with PCC) to reach the vulnerable code path
Exfiltrate sensitive user data
- Successful bypass of entitlement validation (Step 2)
- Apple Intelligence has processed user data (not a freshly wiped device)
- Impact is confined to a single device — no lateral movement or fleet-wide compromise
- Data scope is limited to what Apple Intelligence has processed, not arbitrary filesystem access
- iOS network monitoring and app transport security may flag unusual outbound connections
allowAppleIntelligence = false) on executive and privileged-user devices. This eliminates the vulnerable component entirely. The setting is available in supervised mode on managed devices.- Network-level blocking (firewall/proxy rules) — the attack vector is not tied to a specific network endpoint or domain that can be blocked. Apple Intelligence communicates with PCC over encrypted channels to Apple infrastructure; you cannot selectively block the vulnerable code path without disabling all Apple Intelligence functionality.
- Mobile Threat Defense (MTD) agents — no MTD vendor has published a detection signature for this specific CVE as of 2026-10-28. MTD may detect anomalous behavior post-exploitation but cannot prevent the entitlement bypass itself.
- VPN or ZTNA — routing device traffic through a VPN does not address an on-device authorization flaw. The vulnerability is in how iOS validates entitlements locally, not in network transport.
The supporting signals.
| In-the-Wild Status | No evidence of exploitation. Not listed in CISA KEV. Apple's advisory states no awareness of active exploitation. |
|---|---|
| PoC Availability | None found. No PoC on pocindex.io, GitHub, ExploitDB, or Nuclei templates as of 2026-10-28. No repos named after CVE-2026-86987 exist on GitHub. |
| EPSS | 0.00192 (0.19% probability of exploitation in next 30 days), 8.7th percentile — well below the median. Low predicted exploitation likelihood. |
| KEV Status | Not listed in CISA Known Exploited Vulnerabilities catalog. |
| CVSS Vector (estimated) | CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:L/I:N/A:N → 5.3 (MEDIUM). Network-accessible, no auth, but confidentiality-only with limited scope. NVD has not published an official score; VulDB assigned 4.3. |
| Affected Versions | iOS < 27.2, iPadOS < 27.2 — only on Apple Intelligence-capable hardware: iPhone 15 Pro, iPhone 15 Pro Max, iPhone 16 family, iPhone 17 family; iPad Pro 12.9" 5th gen+, iPad Pro 11" 2nd gen+, iPad Air 5th gen+, iPad mini 7th gen. |
| Fixed Version | iOS 27.2 and iPadOS 27.2, released 2026-10-26. |
| Scanning/Exposure Data | Individual iOS devices are not indexed by Shodan/Censys. Fleet exposure is measurable only through MDM inventory (Jamf, Intune, Kandji, Mosyle). Query for model + OS version. |
| Disclosure Date | 2026-10-26 (coordinated disclosure, patch-on-advisory). |
| Reporting Researcher | Richard Hyunho Im (@richeeta) at Route Zero Security. 21+ Apple acknowledgments, 7+ CVEs, DEF CON 33 Main Stage speaker, Top-25 OpenAI Bug Bounty 2025. Current research focus: reverse-engineering Siri and Apple Intelligence on iOS/macOS 27. |
Sources.
- Apple Security Advisory — iOS 27.2 / iPadOS 27.2
- CyberSecurityNews — iOS 27.2 163 Vulnerabilities
- Strix.ai — CVE-2026-86987 Details
- MacRumors — iOS 27.2 Security Fixes
- Richard Hyunho Im (@richeeta) — Researcher Profile
- HKCERT — Apple Products Multiple Vulnerabilities
- 9to5Mac — 329 Vulnerabilities Across Apple Updates
Why this verdict
- Remote, unauthenticated attack vector sets the floor at MEDIUM. Apple's description says 'a remote attacker' without qualification — implying no authentication is required. Network-reachable + no-auth + data disclosure = at least MEDIUM baseline even before friction adjustments.
- Confidentiality-only impact caps the ceiling. The advisory describes access to 'sensitive user data' with no mention of code execution, integrity modification, or denial of service. This is a read-primitive, not a write-primitive or RCE.
- Narrow device population limits blast radius. Apple Intelligence requires iPhone 15 Pro or later, plus explicit user opt-in. In a typical enterprise of 10,000 managed iPhones, roughly 30-40% may be eligible hardware, and a subset of those will have Apple Intelligence enabled.
- Role multiplier: mobile endpoint. iOS/iPadOS devices are consumer/employee endpoints, not infrastructure. (a) *Low-value role:* personal devices — host-level impact only. (b) *Typical role:* corporate-managed iPhone with email, Slack, Teams — host-level data exposure of one user's AI-processed content. (c) *High-value role:* executive device (CEO/CFO/CISO phone) — the data at stake is genuinely sensitive (board emails, M&A discussions, strategy docs summarized by Apple Intelligence), but the blast radius remains single-host — no domain takeover, no fleet compromise, no lateral movement. iOS is NOT a canonical high-value-role component (not a DC, IdP, hypervisor, CA, or network edge). The high-value-role floor of HIGH does not apply.
- Zero exploitation evidence provides no upward pressure. No PoC, no in-the-wild activity, EPSS at 8.7th percentile, not KEV-listed. The vulnerability is 2 days old with no sign of weaponization.
Why not higher?
The blast radius is confined to a single mobile device's Apple Intelligence data — there is no path to lateral movement, domain compromise, or fleet-scale impact. iOS devices are not canonical high-value infrastructure; even on an executive's phone, the outcome is data disclosure from one endpoint, not organizational takeover. The absence of any PoC, exploitation evidence, or even a detailed technical writeup means the practical risk remains theoretical. Upgrading to HIGH would require either demonstrated exploitation or a chain to code execution, neither of which exists.
Why not lower?
The remote, unauthenticated attack vector with no user interaction required is a meaningful threat — it means any network-reachable device is theoretically exploitable without social engineering. The data at stake (emails, messages, photos processed by Apple Intelligence) is genuinely sensitive, especially on executive or privileged-user devices. Dropping to LOW would understate the confidentiality impact and the frictionless attack prerequisites. VulDB's 4.3 is defensible but slightly underweights the no-auth, no-interaction posture.
Crowdsourced verification payload.
Run this script on a macOS management workstation with a target iOS/iPadOS device connected via USB, or adapt the MDM query section for fleet-wide checks. Requires libimobiledevice (brew install libimobiledevice). Invoke: bash check_cve_2026_86987.sh. No special privileges needed beyond USB device access.
#!/bin/bash
# CVE-2026-86987 Verification Script
# Checks whether a connected iOS/iPadOS device is patched
# Target: Apple Intelligence-capable devices (iPhone 15 Pro+, eligible iPads)
# Exit codes: 0=PATCHED, 1=VULNERABLE, 2=UNKNOWN
FIXED_VERSION="27.2"
CVE_ID="CVE-2026-86987"
COMPONENT="Apple Intelligence"
# Apple Intelligence-capable device identifiers (partial list)
AI_MODELS="iPhone15,3|iPhone15,4|iPhone16|iPhone17|iPad13,4|iPad13,5|iPad13,6|iPad13,7|iPad14|iPad15|iPad16"
version_gte() {
[ "$(printf '%s\n' "$1" "$2" | sort -V | head -n1)" = "$2" ]
}
echo "=== $CVE_ID Check ($COMPONENT) ==="
echo ""
if ! command -v ideviceinfo &>/dev/null; then
echo "[!] ideviceinfo not found. Install: brew install libimobiledevice"
echo ""
echo "For MDM fleet checks, query your MDM for:"
echo " Model IN (iPhone 15 Pro+, eligible iPads) AND OS < $FIXED_VERSION"
echo ""
echo "UNKNOWN - tooling not available"
exit 2
fi
PRODUCT_VERSION=$(ideviceinfo -k ProductVersion 2>/dev/null)
PRODUCT_TYPE=$(ideviceinfo -k ProductType 2>/dev/null)
DEVICE_NAME=$(ideviceinfo -k DeviceName 2>/dev/null)
if [ -z "$PRODUCT_VERSION" ]; then
echo "UNKNOWN - No device connected or ideviceinfo failed"
exit 2
fi
echo "Device: $DEVICE_NAME ($PRODUCT_TYPE)"
echo "OS: $PRODUCT_VERSION"
echo "Fixed: >= $FIXED_VERSION"
echo ""
# Check if device supports Apple Intelligence
if echo "$PRODUCT_TYPE" | grep -qE "$AI_MODELS"; then
echo "[*] Device model supports Apple Intelligence"
else
echo "[*] Device model may NOT support Apple Intelligence"
echo " If Apple Intelligence is not available, device is NOT AFFECTED"
echo ""
echo "PATCHED - device not eligible for Apple Intelligence"
exit 0
fi
echo ""
if version_gte "$PRODUCT_VERSION" "$FIXED_VERSION"; then
echo "PATCHED - running $PRODUCT_VERSION (>= $FIXED_VERSION)"
exit 0
else
echo "VULNERABLE - running $PRODUCT_VERSION (< $FIXED_VERSION)"
echo " Action: Update to iOS/iPadOS $FIXED_VERSION or later"
echo " Workaround: Disable Apple Intelligence via MDM profile"
exit 1
fi