CVE-2026-49496: Ghidra Heap-Use-After-Free in Sleigh Decompiler—Patch v12.1

Technical summary

CVE-2026-49496 is a heap-use-after-free vulnerability (CWE-416) in the SleighBuilder::generatePointerAdd function within Ghidra's Sleigh decompilation library. The root cause is iterator invalidation: when PcodeCacher::allocateInstruction reallocates the internal 'issued' vector during instruction caching, existing iterators pointing into that vector become invalid. Subsequent pointer arithmetic operations in SleighBuilder then dereference stale pointers, corrupting heap metadata or live objects. The vulnerability is exposed through the public Sleigh::oneInstruction C++ API, making any downstream application or library that consumes the SLEIGH library potentially affected. Exploitation does not require elevated privileges and occurs purely through decompilation of a crafted binary.

Business impact

Organizations using Ghidra or embedding the SLEIGH library in security analysis pipelines face denial-of-service risk: a malicious binary could crash the analysis tool, disrupting incident response, malware triage, or software analysis workflows. The memory corruption aspect introduces uncertainty about tool reliability—analysts may question whether results from a Ghidra session on a suspicious file are trustworthy. For teams running Ghidra in automated reverse-engineering or threat intelligence environments, this becomes a stability concern. The attack surface is limited to users who explicitly open untrusted binaries, so risk is proportional to your organization's handling of third-party or suspected-malicious executables.

Affected systems

Ghidra versions prior to 12.1 are affected. This includes all 12.0.x releases and earlier versions. The vulnerability specifically impacts the public Sleigh::oneInstruction C++ API, so affected parties include: (1) direct Ghidra users decompiling binaries via the GUI or command-line tools, and (2) downstream developers who have integrated the SLEIGH library as a dependency in custom reverse-engineering, binary analysis, or threat intelligence tools. Any tool statically or dynamically linking against vulnerable Ghidra/SLEIGH libraries is in scope.

Exploitability

Exploitability is practical but not trivial. An attacker must craft a malicious ELF, PE, or other binary format that, when parsed and decompiled by Ghidra, triggers the specific reallocation pattern in PcodeCacher that invalidates the iterator. This requires knowledge of Sleigh's internal caching behavior and the ability to engineer a binary that stresses the instruction cache in the right way. The attack requires user interaction—someone must open the malicious binary in Ghidra—making it suitable for targeted attacks against security researchers, analysts, or automated analysis pipelines that ingest untrusted files. It is not a network-exploitable vector and does not grant remote code execution by itself, though heap corruption could theoretically be leveraged for further exploitation in a chain.

Remediation

Upgrade Ghidra to version 12.1 or later. NSA has released fixes addressing the iterator invalidation in PcodeCacher and ensuring that SleighBuilder does not use invalidated iterators. Users unable to upgrade immediately should avoid decompiling untrusted or suspicious binaries until patching is complete, or run Ghidra in an isolated sandbox environment. For organizations embedding SLEIGH in custom tools, verify that your dependency management pulls in the patched version and rebuild/redeploy those applications.

Patch guidance

Visit the official NSA Ghidra release page (https://github.com/NationalSecurityAgency/ghidra/releases) and download Ghidra 12.1 or later. The patch is available as a binary release and source archive. For package manager distributions (Linux), check your distro's package repositories for Ghidra 12.1+. If you maintain a custom build or have integrated SLEIGH as a library, update your source dependency to the 12.1 tag or later commit. Apply the patch during your next maintenance window; the severity (CVSS 6.1 MEDIUM) does not require emergency out-of-band patching but should be prioritized within 30 days.

Detection guidance

On systems running Ghidra, verify the installed version (Help > About in the GUI, or `ghidra --version` on CLI) and confirm it is 12.1 or later. For custom tools embedding SLEIGH, check dependency manifests (pom.xml, CMakeLists.txt, build.gradle, etc.) and confirm the Ghidra/SLEIGH version specified is 12.1+. Monitor Ghidra process crashes or unusual memory behavior when decompiling files from untrusted sources; a crash with a heap corruption signature (e.g., malloc/free errors) could indicate exploitation attempts. Enable core dumps and examine heap allocator diagnostics if available.

Why prioritize this

This vulnerability should be patched within 30 days. The CVSS 6.1 MEDIUM rating reflects moderate risk: local attack surface, user interaction required, and no confidentiality impact. However, the memory corruption nature introduces tool reliability uncertainty, and the public API exposure means downstream libraries are also affected. Prioritize patching if your organization runs Ghidra against untrusted binaries regularly (threat research, malware analysis, third-party software vetting). If Ghidra is used only on internally-developed or trusted code, risk is lower, but patching remains prudent for long-term stability.

Risk score, explained

CVSS 3.1 score of 6.1 (MEDIUM) reflects: Attack Vector Local (malicious binary must be opened locally), Attack Complexity Low (once a suitable binary is crafted, the trigger is straightforward), Privileges Required None (no special privileges needed to run Ghidra), User Interaction Required (user must open the binary), Scope Unchanged (impact is limited to Ghidra process), Confidentiality None (no data disclosure), Integrity Low (heap corruption, potential for minor data corruption), Availability High (process crash is likely). The 'High' availability impact drives the score above 5.0; the requirement for local user interaction and lack of escalation to confidentiality loss prevents a higher score.

Frequently asked questions