The zlib maintainers released version 1.4 on November 4, repairing memory errors in the compression library and flaws in its bundled minizip archive tools. Developers need to check which functions their applications call and which library copies their releases actually ship.
The maintainers’ release notes and changelog describe the fixes and remaining limitations. Michael Larabel’s November 5 Phoronix report summarizes the release. A later discussion-site submission does not change those publication dates.
Evidence status: published security fixes and documented trigger conditions. The reviewed release notes and reporting provide no evidence of active exploitation or a measured count of compromised applications.
Different functions create different exposure
zlib compresses and decompresses data inside other programs. Its presence in a dependency list does not show that an application reaches every affected function.
- Non-blocking gzip writes: CVE-2026-85091 concerns memory corruption in a particular writing path. The published CVE record lists 1.3.1.2 through 1.3.2 as affected. zlib identifies 1.4 as containing the repair. This is not evidence that every ordinary decompression call has the same exposure.
- Dictionary retrieval: the maintainers say
inflateGetDictionary()could write up to 8 MB beyond a caller’s 32K buffer under specific invalid-input and state-reset conditions. They identify 1.3.1 and 1.3.2 as affected by that issue. - Other memory handling: 1.4 repairs additional out-of-bounds accesses and cases where uninitialized memory could reach output. The notes describe separate conditions involving stream synchronization, bit priming, seeking and resets. They do not supply one affected-version range covering every fix.
- Archive handling: the bundled minizip code receives memory-management and archive-validation fixes. Its
miniunzextraction tool could create directories outside the extraction directory and, on Windows, write files elsewhere. Exposure depends on using that code to process an attacker’s archive.
For an application accepting untrusted compressed files or archives, establish which reader or extractor is actually used. A product using a different ZIP library should not inherit minizip’s findings merely because it also links zlib.
Update the shipped component, not just the host package
For operators, use supported operating-system and application security updates. Check the supplier’s advisory for backported fixes; a distributor may repair an older version without adopting upstream’s full release number.
For developers, inventory vendored source, static libraries and container images as well as the operating system’s shared library. Rebuild and redeploy applications that embed affected copies. Updating a host package cannot replace code already compiled into a binary or copied into an image.
Test the upgrade against real application behavior. The maintainers flag changed CMake library names on MSVC, a minimum CMake version of 3.15, and changes to offset types. They specifically tell Windows applications built with 1.3.2 headers to recompile.
These are release-engineering requirements, not a reason to postpone affected components indefinitely. Verify that the rebuilt artifact uses the intended repaired component and passes normal compression, extraction and error-handling tests.
The release retains an explicit unsafe-input warning
Version 1.4’s known-issues section says contrib/iostream2 remains unsafe on untrusted input. A length read from compressed data can exceed the destination buffer. The maintainers recommend contrib/iostream3 instead.
The notes also discuss CVE-2026-4738 in contrib/infback9. They say the corrected pointer arithmetic returns in bounds before use and no known compiler mishandles it. They caution that its critical rating overstates the demonstrated behavior, while noting possible traps on capability hardware such as CHERI.
That qualification belongs beside the identifier. A severity label cannot substitute for checking the affected component, reachable conditions and observed outcome.
Caveat: Upstream 1.4 fixes the listed issues but does not make every contributed component safe for hostile input; application exposure and distributor backports require separate checks.