← Vulnerability feed

Vulnerability record · CVE-2026-85670 · published 4 September 2026

CVE-2026-85670: Out-of-bounds write vulnerability

tokenizers (Hugging Face) is affected by an out-of-bounds buffer access in BpeBuilder::build (tokenizers/src/models/bpe/model.rs). When loading a tokenizer.json via Tokenizer::from_file/from_str, the builder sizes a scratch buffer to the longest vocabulary key, then writes each concatenated merge rule into it. A merge whose concatenated token exceeds the longest vocabulary key overruns the buffer, which Rust turns into a panic that aborts the process in Rust and FFI embeddings. This occurs at load time with no encoding required, so an attacker who supplies a crafted tokenizer.json can cause a denial of service. A secondary defect at the same location can cause a usize underflow (panic in debug, potential memory corruption in release) when continuing_subword_prefix is set and a merge token is shorter than the prefix. Observed in version 0.23.1.

7.1 CVSS 4.0 High EPSS 0.47% · top 62.1% CWE-787 · Out-of-bounds write Awaiting Analysis
7.1CVSS 4.0 base score
0.47%EPSS exploitation probability, 30 days
NoNot in CISA KEV
0Affected product versions listed by NVD
5References
24 Sep 2026Last modified by NVD

Description

tokenizers (Hugging Face) is affected by an out-of-bounds buffer access in BpeBuilder::build (tokenizers/src/models/bpe/model.rs). When loading a tokenizer.json via Tokenizer::from_file/from_str, the builder sizes a scratch buffer to the longest vocabulary key, then writes each concatenated merge rule into it. A merge whose concatenated token exceeds the longest vocabulary key overruns the buffer, which Rust turns into a panic that aborts the process in Rust and FFI embeddings. This occurs at load time with no encoding required, so an attacker who supplies a crafted tokenizer.json can cause a denial of service. A secondary defect at the same location can cause a usize underflow (panic in debug, potential memory corruption in release) when continuing_subword_prefix is set and a merge token is shorter than the prefix. Observed in version 0.23.1.

CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:P/VC:N/VI:N/VA:H/SC:N/SI:N/SA:N/E:X/CR:X/IR:X/AR:X/MAV:X/MAC:X/MAT:X/MPR:X/MUI:X/MVC:X/MVI:X/MVA:X/MSC:X/MSI:X/MSA:X/S:X/AU:X/R:

References

LinkTags
https://github.com/huggingface/tokenizers
https://github.com/huggingface/tokenizers/blob/v0.23.2/tokenizers/src/models/bpe/model.rs
https://github.com/huggingface/tokenizers/issues/2094
https://www.vulncheck.com/advisories/tokenizers-bpebuilder-buffer-overflow-via-merge-token
https://github.com/huggingface/tokenizers/issues/2094

Track CVE-2026-85670 inside VULONE

Watch it alongside the ransomware crews, C2 infrastructure and forum chatter that reference it, query it through the API and pull it into your SIEM over TAXII.

Start free Open in platform

Related vulnerabilities

Same products first, then exploited flaws of the same weakness class.

8.8CVE-2026-53266Linux kernel ebtables SNAT out-of-bounds write in ARP rewriteThe ebtables SNAT target rewrites the ARP sender hardware address via skb_store_bits() without first making that range writable. When the ARP SHA byt…KEVEPSS 0.65%analysed8.8CVE-2026-87491Google Chrome V8 out-of-bounds write enables sandbox code executionChrome before 153.0.8010.36 contains an out-of-bounds write in the V8 JavaScript engine. A crafted HTML page can trigger the memory corruption, and b…KEVEPSS 3.1%analysed9.8CVE-2025-25249Fortinet FortiOS and FortiSwitchManager heap buffer overflow via crafted packetsA heap-based buffer overflow (CWE-122/CWE-787) in Fortinet FortiOS 6.4 through 7.6.3 and FortiSwitchManager 7.0 through 7.2.6 lets an unauthenticated…KEVEPSS 3.9%analysed7.8CVE-2026-53362Linux kernel IPv6 UDP paged allocation out-of-bounds write__ip6_append_data() in the Linux kernel mis-accounts fraggap on the paged-allocation path, leaving the linear skb area undersized while pagedlen is o…KEVEPSS 0.71%analysed7.8CVE-2022-0995Linux kernel watch_queue out-of-bounds writeThe Linux kernel's watch_queue event notification subsystem contains an out-of-bounds write (CWE-787) that can overwrite kernel state. A local user c…KEVEPSS 8.8%analysed8.8CVE-2026-11645Google Chrome V8 out-of-bounds read and write enables sandbox code executionGoogle Chrome before 149.0.7827.103 contains an out-of-bounds read and write in the V8 JavaScript engine. A crafted HTML page can trigger the memory …KEVEPSS 2.2%analysed9.3CVE-2026-0300PAN-OS User-ID Authentication Portal buffer overflow allows root code executionA buffer overflow (out-of-bounds write, CWE-787) in the User-ID Authentication Portal (Captive Portal) service of Palo Alto Networks PAN-OS lets an u…KEVEPSS 32%analysed8.8CVE-2026-3909Google Chrome Skia out-of-bounds write via crafted HTML pageChrome's Skia graphics library contains an out-of-bounds write (CWE-787) that a remote attacker can trigger with a crafted HTML page. It is fixed in …KEVEPSS 0.70%analysed

Source: NIST National Vulnerability Database (record CVE-2026-85670), CISA KEV, FIRST EPSS (scores of 2026-09-26). This page is refreshed as NVD updates the record.