Loading vulnerability…
CVE-2026-64747 is a buffer overflow vulnerability in the AVEVideoEncoder component of Apple operating systems. The flaw was addressed through improved size validation, indicating that insufficient bounds checking on attacker-controlled or malformed input could lead to memory corruption. Successful exploitation may allow a local application to execute arbitrary code in kernel context. The issue affects multiple Apple platforms and was fixed in iOS 26.6, iPadOS 26.6, macOS Sequoia 15.7.8, macOS Sonoma 14.8.8, macOS Tahoe 26.6, tvOS 26.6, visionOS 26.6, and watchOS 26.6.
Mallory correlates every CVE against your assets, your vendors, and active adversary campaigns. Know which vulnerabilities matter for you, not just which ones are loud.
What it means. What to do now. Patch path, mitigations, and the assume-compromise checklist.
What an attacker gets, and what they’ve been doing with it.
If you can’t patch tonight, do this now.
Patch, then assume compromise.
1 valid exploit after Mallory filtered fakes, detection scripts, and README-only repos.
This repository is a standalone Swift/Xcode iOS application, not part of a common exploit framework. Its purpose is to probe and potentially trigger CVE-2026-64747, described as an Apple AGXG14P command queue count-bitmask out-of-bounds condition on iPhone 13/A15 running iOS 26.5.2. The repo contains 12 files total, with the main logic in four Swift source files: AGXProbeApp.swift (app entry), ContentView.swift (UI/controller), AGXStress.swift (GPU stress/trigger logic), and AGXDump.swift (userland binary dumping). There is also an Xcode project, Info.plist, README, GitHub Actions build workflow, and a large repos.json metadata file. The main exploit capability is in AGXStress.swift. It creates a Metal device and command queue, builds inline Metal shader code at runtime, and repeatedly exercises multiple GPU submission patterns intended to influence the vulnerable queue count bitmask path. The documented phases include slotStorm, eventStorm, bitStorm, maskStorm, hugeBatch, inFlight, fenceStorm, renderPath, multiQueue, computeHeavy, chainedEvents, and mixed. These phases attempt to drive high-bit values, large batches, many in-flight command buffers, fences, render/compute/blit activity, and multi-queue pressure. The expected outcome is not code execution but empirical triggering of the kernel bug, evidenced by app death, GPU stall, or full device reboot/kernel panic. Because no post-exploitation payload or shell is included, this is best classified as a proof-of-concept trigger probe rather than a weaponized exploit. A secondary capability is AGXDump.swift, which enumerates dyld-loaded images in the current process and dumps binaries whose paths contain AGX, Metal, CoreMTL, IOKit, or IOGPU. It reconstructs rebased memory images from mapped segments and writes them to the app Documents directory under agx_dump, along with metadata text files containing segment and symbol table information. This supports reverse engineering of the userland side of the vulnerability without jailbreak access. No network exploitation or C2 behavior is present. The observable endpoints are almost entirely local file paths and artifact names: the app writes agxprobe.log and agx_dump outputs into the Documents directory, exposes them via UIFileSharingEnabled, and the CI workflow builds AGXProbe.app and packages AGXProbe.ipa. The README references sideloading/install URLs and operational guidance, but the exploit logic itself does not contact remote hosts. Overall, the repository is a local, device-resident iOS GPU vulnerability trigger harness plus binary-dumping utility for Apple AGX/Metal reverse engineering. It is a real exploit-related PoC, but not a full exploit chain and not a detection-only script.
Products and vendors Mallory has correlated with this vulnerability. Open in Mallory to drill down to specific CPE configurations and version ranges.
Vendor-confirmed product mapping. Mallory continuously reconciles this list against your asset inventory.
3 sources tracked across advisories, community write-ups, and news. New activity surfaces here as Mallory finds it.
Query your assets running an affected version, and investigate the blast radius.
Every observed campaign linking this CVE to a named adversary.
Malware families riding this exploit, with evidence and IOCs.
YARA, Sigma, Snort, and vendor rules, auto-deployed to your SIEM.
Cross-references every affected SKU, including bundled OEM variants.
Community discussion across Reddit, Mastodon, and other social sources.