Critical

Heap Buffer Overflow in llama.cpp KV Cache State Restore

IdentifiersCVE-2026-43629CWE-787· Out-of-bounds Write

CVE-2026-43629 is a heap buffer overflow vulnerability in llama.cpp affecting builds b4882 through b9058. The flaw is in the KV cache state restore path, specifically in the state_read_data() function, which computes the amount of data to write without proper overflow validation. A crafted state file can cause the cell_count multiplication to overflow or otherwise exceed the size of the allocated tensor buffer, resulting in attacker-controlled data being written past heap buffer boundaries during state restoration. This memory corruption condition can affect heap metadata and in-memory model structures and may be exploitable for arbitrary code execution, including through corruption of function pointers.

For your environment

Are you exposed to this one?

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.

Start free trial

ANALYST BRIEF

Impact, mitigation & remediation

What it means. What to do now. Patch path, mitigations, and the assume-compromise checklist.

Impact

What an attacker gets, and what they’ve been doing with it.

Successful exploitation can corrupt heap memory during KV cache state restoration. Documented consequences include heap metadata corruption, model weight corruption, process instability, and denial of service. In the worst case, the overflow may be leveraged for arbitrary code execution through overwrite of sensitive in-memory structures such as function pointers.

Mitigation

If you can’t patch tonight, do this now.

Restrict write access to the slot_save_path directory to trusted local users and processes only. Do not restore untrusted, externally supplied, or tampered KV cache state files until a fixed build is deployed. Where possible, disable or avoid use of the state restore workflow in environments that cannot guarantee integrity of saved state data.

Remediation

Patch, then assume compromise.

Upgrade llama.cpp to a fixed build newer than b9058 that adds proper bounds and overflow checking in the KV cache state restore logic. The vulnerable size calculations in state_read_data() should be corrected to validate multiplication results and ensure the computed write length cannot exceed the destination tensor buffer allocation before any copy or restore operation proceeds.

PUBLIC EXPLOITS

Exploits

No valid public exploits. Mallory filtered out 1 candidate as fakes, detection scripts, or README-only repos.

VALID 0 / 1 TOTALView more in app

All candidate exploits were filtered out by Mallory's validation.

EXPOSURE SURFACE

Affected products & vendors

Products and vendors Mallory has correlated with this vulnerability. Open in Mallory to drill down to specific CPE configurations and version ranges.

VendorProductType

GgmlLlama.Cppapplication

ggml-orgLlama.Cppapplication

Vendor-confirmed product mapping. Mallory continuously reconciles this list against your asset inventory.

ACTIVITY FEED

Recent activity

5 sources tracked across advisories, community write-ups, and news. New activity surfaces here as Mallory finds it.

5 SOURCESView more in app

threataft blogNews

Aug 7, 2026

llama.cpp 6-CVE Bundle - Multiple Memory Corruption Vulnerabilities (CVSS up to 9.2) | ThreatAft

A critical heap buffer overflow in llama.cpp's KV cache state restore path that can enable heap corruption, model weight corruption, or arbitrary code execution.

What this page doesn’t show

The version that knows your environment.

This page is what’s public. Mallory adds the parts that aren’t: which of your assets are affected, which adversaries are exploiting it right now, which detections to deploy, and what to do tonight.

Start free trial

Exposure mapping

Query your assets running an affected version, and investigate the blast radius.

Threat actor evidence

Every observed campaign linking this CVE to a named adversary.

Associated malware

Malware families riding this exploit, with evidence and IOCs.

Detection signatures

YARA, Sigma, Snort, and vendor rules, auto-deployed to your SIEM.

Vendor-by-vendor mapping

Cross-references every affected SKU, including bundled OEM variants.

Social activity4