Apple ImageIO CVE-2026-65346: the risk inside image decoding

Before an image appears on screen, software has to parse its structure, decompress its contents and allocate memory for the result. That work can also happen while generating a thumbnail or reading metadata. Apple’s August ImageIO fix concerns the arithmetic involved in processing that untrusted input.

The confirmed issue and fixed releases

Apple describes an integer overflow in ImageIO with a potential arbitrary-code-execution impact during image processing. Its stated correction is stronger input validation. The August 17 advisories list the fix in iOS and iPadOS 26.6.1 and macOS Tahoe 26.6.2. These are the releases associated with this disclosure, not a complete inventory of every Apple platform.

Apple’s iOS and iPadOS 26.6.1 advisory

Apple’s macOS Tahoe 26.6.2 advisory

What happens before the picture appears

Apple’s CGImageSource interface reads image data, manages buffers, decompresses supported formats, and exposes thumbnails and metadata. Those tasks can exercise a decoder without a separate photo-editing session. They explain the framework’s role, but do not identify how this particular CVE was reached.

Apple developer documentation: CGImageSource

Image decoders calculate dimensions, channel counts, row strides and storage requirements. If arithmetic exceeds the range of its representation, the result may no longer describe the intended size. Later code that still relies on the original assumption can then get memory bounds wrong. That is a general explanation of integer overflow; Apple has not identified the calculation that failed here.

MITRE’s definition of integer overflow, CWE-190

Recognizing a file format is only the first check. A valid signature does not guarantee that lengths, dimensions or relationships between values will stay safe throughout decoding. Applications accepting images also need resource limits and predictable failure handling, even when a system library does the decoding.

The details Apple has not published

The cited advisories do not name an image format, vulnerable function, delivery path or observed attack campaign. They also do not establish a sandbox escape or kernel privileges. Arbitrary code execution is a serious possible outcome in the affected process; that process’s privileges and isolation still determine what else is reachable.

We can explain the disclosed component and integer-overflow class without assuming that every received image can silently take over an iPhone. The public record does not support that broader claim.

Check the system library’s update

For this issue, record the operating-system branch and installed version. Updating only a photo editor does not demonstrate that the underlying operating-system ImageIO fix has been installed. Apple’s current release index lists iOS and iPadOS 26.6.2 as available at our source check; install the latest compatible supported release offered for the device, using the relevant advisory to confirm coverage.

Apple’s current security-release index

Include shared intake workstations and Macs that process images automatically when reviewing updates. After maintenance, record the installed OS and check a normal import or thumbnail operation with a harmless sample. A suspicious image should not be used to test whether patching worked.

Developers should identify where their applications call system decoders and which process handles the result. Record the input-size limits, error behavior and service account’s access. The Apple update corrects this disclosed flaw; those design choices help limit the consequences when another decoder bug turns up.

Source check: September 8, 2026. This is Short Circuit LLC’s independent analysis of public disclosures and documented architecture. Apple credits Meta Red Team X researcher Nik Tsytsarkin with CVE-2026-65346.