CVE-2026-85670

MEDIUMPre-NVD 6.56.5
EchelonGraph scoreMEDIUM confidence

“—” beside the score: it is equal to the CVSS score beside it.

Score 6.5 from GitHub Security Advisory (severity: HIGH) published 2026-09-04. A secondary CVSS source baseline 6.5; sources differ by 0.0.

Triggered by: GitHub Security Advisory CVSS
Sources: EPSS, GitHub Security Advisory, a secondary CVSS source
Elevated
6.5EG
EchelonGraph verdictPlan mitigationRemediate within your standard update timelines (the SSVC decision on this page).
  • CISA's Vulnrichment rates exploitation as PoC; EchelonGraph holds no exploit-code reference for it
CISA-KEV: Not listedEPSS PROB: 0.5%CVSS: 6.5Exploit: No code on recordExposed services: Not assessed

No fix is confirmed yet — apply a workaround or compensating control (WAF / firewall / segmentation) and watch for the fix.

The CVE record's CNA (VulnCheck) lists huggingface tokenizers versions up to and including 0.23.2 as affected. No release containing a fix is confirmed on record.

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 v3
6.5
EG Score
6.5MEDIUMmedium confidence
EG Risk
45
EG Risk 45/100

EG Risk is EchelonGraph's 0–100 priority score. It fuses impact severity (EG Severity, the CVSS base score, never the KEV-floored or EPSS-lifted EG Score) with real-world exploitation and automatability, under weights EchelonGraph set by hand, so you can rank equal-severity CVEs and fix the most dangerous first. Higher = act sooner.

How it’s computed (value × EchelonGraph’s weight)
Severity65% × 45%
Impact severity (EG Severity): CVSS 6.5 of 10. Never the KEV-floored or EPSS-lifted EG Score.
Exploitation40% × 40%
EchelonGraph's fixed weight for EG-KEV Tier 3 (Elevated): a constant, not a probability.
Automatability0% × 15%
CISA's SSVC Automatable value: no (Vulnrichment).
SSVC (CISA's decision table, applied by EchelonGraph): Track at low or medium mission impact; Track* at high (mission-essential systems).
Exploitation public PoC (CISA Vulnrichment) · Automatable no (CISA Vulnrichment) · Technical impact partial (CISA Vulnrichment). All three are CISA's SSVC values (Vulnrichment). EchelonGraph applied CISA's decision table to them; CISA publishes SSVC inputs and the table, not a decision for each CVE.
Action: No fix is confirmed yet. Restrict network exposure of the affected system or apply the vendor's mitigation within your standard update timelines, and watch the vendor's advisory for the fix.
EPSS PROB
0.5%
EPSS %ILE
39th
KEV
Not listed

What the EG score is made of (scorer v4.0: the three questions the 0–10 score mixes, published apart · methodology)

The EG score (6.5): severity, escalated by confirmed exploitation. This CVE's 6.5 is its impact severity (EG Severity), not escalated.

EG Severity
6.5 MEDIUMCVSS-B, impact only: no KEV listing, EPSS or EG-KEV tier raises it.GitHub Security Advisory CVSS
EG Likelihood
0.5% within 30 daysSource: FIRST EPSS (model v2026.06.15, score date 2026-10-17). FIRST's exploitation probability, not an EchelonGraph model.
Confirmed exploitation
None on record (CISA KEV, EG-KEV Tier 1)Reported apart from the likelihood: an observed fact, not a probability.
EG Priority
SSVC Track at medium mission impactTrack at low or medium mission impact; Track* at high (mission-essential systems).CISA BOD 26-04: 60 days if the affected system is publicly exposed; Fix on system upgrade if it is not.Each input and its source: the SSVC and BOD 26-04 blocks below.

SSVC (CISA's decision table, applied by EchelonGraph)Track at low or medium mission impact; Track* at high (mission-essential systems).

No fix is confirmed yet. Restrict network exposure of the affected system or apply the vendor's mitigation within your standard update timelines, and watch the vendor's advisory for the fix.

Exploitation public PoC (CISA Vulnrichment) · Automatable no (CISA Vulnrichment) · Technical impact partial (CISA Vulnrichment). All three are CISA's SSVC values (Vulnrichment). EchelonGraph applied CISA's decision table to them; CISA publishes SSVC inputs and the table, not a decision for each CVE. Mission impact is CISA's Mission & Well-being decision point, and only you can judge it. CISA's table makes it high when Mission Prevalence is Essential — the vulnerable component "directly provides capabilities that constitute at least one MEF for at least one entity" (MEF: mission essential function) — or when Public Well-Being Impact is Irreversible: "multiple fatalities are likely", the cyber-physical system "is likely lost or destroyed", "extreme or serious externalities" are imposed on other parties, or social systems such as elections or the financial grid "are destabilized and potentially collapse". CISA's decision table

BOD 26-04 (CISA's remediation timeline, Table 1)60 days if the affected system is publicly exposed; Fix on system upgrade if it is not.

In CISA KEV no (CISA's KEV catalog) · Automatable no (CISA Vulnrichment) · Technical impact partial (CISA Vulnrichment). Whether your system is publicly exposed is yours to answer. BOD 26-04 binds US Federal Civilian Executive Branch agencies; for anyone else it is CISA's published timeline, a reference rather than an obligation. The directive · how EchelonGraph applies it

Published

September 4, 2026

Last Modified

September 24, 2026

Advisory Details (4)

Auto-updated Sep 4, 2026
No fix on record yet.
Web page

tokenizers BpeBuilder Buffer Overflow via merge token | Advisories | VulnCheck

https://www.vulncheck.com/advisories/tokenizers-bpebuilder-buffer-overflow-via-merge-token
Web page

tokenizers/tokenizers/src/models/bpe/model.rs at v0.23.2 · huggingface/tokenizers · GitHub

https://github.com/huggingface/tokenizers/blob/v0.23.2/tokenizers/src/models/bpe/model.rs
Web page

GitHub - huggingface/tokenizers: 💥 Fast State-of-the-Art Tokenizers optimized for Research and Production · GitHub

https://github.com/huggingface/tokenizers
Web page

[Security] Load-time process abort when loading a malicious tokenizer.json (BPE merge buffer overrun) · Issue #2094 · huggingface/tokenizers · GitHub

https://github.com/huggingface/tokenizers/issues/2094

Vendor Advisories for CVE-2026-85670(1)

These vendors published their own advisory mentioning this CVE — often with vendor-specific remediation steps + affected product lists not in NVD.

Weakness Classification(1)

MITRE Common Weakness Enumeration — the root-cause categories this CVE belongs to.

Data Freshness Timeline

(refreshed 8× in last 7d / 72× in last 30d)

Each row is a time one of our pipelines fetched, updated or re-scored this CVE. For example, "NVD update" means NVD published or revised its analysis for this CVE; "MITRE cvelistV5" means we ingested or refreshed it from the CNA feed. An arrow marks a change the record states ("CVSS v3 → 9.1"); EPSS and GHSA rows show the values recorded at that time, not what changed (EPSS values are rounded; hover one for the exact figures). Most recent first.

Showing the most recent 98 of 118 refreshes on record for this CVE.

A repeat, a row identical to one the same pipeline wrote less than 10 seconds earlier, is not a separate refresh and is not counted. 2 repeat rows among the latest 100 are merged into the row they repeat. The 20 oldest rows on record are not shown and were not checked, so the counts can still include repeats.

  1. 2026-10-19 16:33 UTCEPSS rescoreEPSS 0.5% · 39th percentile
  2. 2026-10-18 16:33 UTCEPSS rescoreEPSS 0.5% · 39th percentile
  3. 2026-10-18 03:05 UTCEG score recompute
  4. 2026-10-17 16:35 UTCEPSS rescoreEPSS 0.5% · 39th percentile
  5. 2026-10-16 16:44 UTCEPSS rescoreEPSS 0.5% · 39th percentile
  6. 2026-10-15 16:33 UTCEPSS rescoreEPSS 0.5% · 38th percentile
  7. 2026-10-14 16:33 UTCEPSS rescoreEPSS 0.5% · 38th percentile
  8. 2026-10-13 16:35 UTCEPSS rescoreEPSS 0.5% · 38th percentile
  9. 2026-10-12 16:35 UTCEPSS rescoreEPSS 0.5% · 38th percentile
  10. 2026-10-12 14:52 UTCEG score recompute
  11. 2026-10-11 19:17 UTCEG score recompute
  12. 2026-10-11 16:34 UTCEG score recompute
  13. 2026-10-11 13:51 UTCEG score recompute
  14. 2026-10-11 13:49 UTCEPSS rescoreEPSS 0.5% · 38th percentile
  15. 2026-10-10 20:00 UTCEPSS rescoreEPSS 0.5% · 38th percentile
  16. 2026-10-10 00:12 UTCEG score recompute
  17. 2026-10-10 00:12 UTCGHSA enrichmentGHSA-jrf5-f6jp-7vxc · GitHub severity HIGH · GitHub CVSS 6.5
  18. 2026-10-09 20:07 UTCEPSS rescoreEPSS 0.5% · 38th percentile
  19. 2026-10-09 01:14 UTCEG score recompute
  20. 2026-10-09 01:13 UTCGHSA enrichmentGHSA-jrf5-f6jp-7vxc · GitHub severity HIGH · GitHub CVSS 6.5
  21. 2026-10-08 17:13 UTCEPSS rescoreEPSS 0.5% · 38th percentile
  22. 2026-10-07 23:48 UTCEPSS rescoreEPSS 0.5% · 38th percentile
  23. 2026-10-07 23:11 UTCGHSA enrichmentGHSA-jrf5-f6jp-7vxc · GitHub severity HIGH · GitHub CVSS 6.5
  24. 2026-10-06 21:57 UTCEG score recompute
  25. 2026-10-06 21:57 UTCGHSA enrichmentGHSA-jrf5-f6jp-7vxc · GitHub severity HIGH · GitHub CVSS 6.5
Show 73 more
  1. 2026-10-06 15:18 UTCEPSS rescoreEPSS 0.5% · 38th percentile
  2. 2026-10-05 23:01 UTCEG score recompute
  3. 2026-10-05 23:01 UTCGHSA enrichmentGHSA-jrf5-f6jp-7vxc · GitHub severity HIGH · GitHub CVSS 6.5
  4. 2026-10-05 17:44 UTCEPSS rescoreEPSS 0.5% · 38th percentile
  5. 2026-10-04 23:55 UTCEG score recompute
  6. 2026-10-04 23:55 UTCGHSA enrichmentGHSA-jrf5-f6jp-7vxc · GitHub severity HIGH · GitHub CVSS 6.5
  7. 2026-10-04 23:22 UTCEPSS rescoreEPSS 0.5% · 38th percentile
  8. 2026-10-04 00:57 UTCEG score recompute
  9. 2026-10-04 00:57 UTCGHSA enrichmentGHSA-jrf5-f6jp-7vxc · GitHub severity HIGH · GitHub CVSS 6.5
  10. 2026-10-03 14:26 UTCEPSS rescoreEPSS 0.5% · 38th percentile
  11. 2026-10-03 00:22 UTCEG score recompute
  12. 2026-10-03 00:22 UTCGHSA enrichmentGHSA-jrf5-f6jp-7vxc · GitHub severity HIGH · GitHub CVSS 6.5
  13. 2026-10-02 17:57 UTCEPSS rescoreEPSS 0.5% · 38th percentile
  14. 2026-10-02 00:37 UTCEG score recompute
  15. 2026-10-02 00:37 UTCGHSA enrichmentGHSA-jrf5-f6jp-7vxc · GitHub severity HIGH · GitHub CVSS 6.5
  16. 2026-10-01 19:50 UTCEPSS rescoreEPSS 0.5% · 38th percentile
  17. 2026-10-01 01:39 UTCEG score recompute
  18. 2026-10-01 01:39 UTCGHSA enrichmentGHSA-jrf5-f6jp-7vxc · GitHub severity HIGH · GitHub CVSS 6.5
  19. 2026-09-30 15:04 UTCEPSS rescoreEPSS 0.5% · 38th percentile
  20. 2026-09-30 02:43 UTCEG score recompute
  21. 2026-09-30 02:43 UTCGHSA enrichmentGHSA-jrf5-f6jp-7vxc · GitHub severity HIGH · GitHub CVSS 6.5
  22. 2026-09-28 23:57 UTCEG score recompute
  23. 2026-09-28 23:57 UTCGHSA enrichmentGHSA-jrf5-f6jp-7vxc · GitHub severity HIGH · GitHub CVSS 6.5
  24. 2026-09-28 13:52 UTCEPSS rescoreEPSS 0.5% · 37th percentile
  25. 2026-09-28 01:00 UTCEG score recompute
  26. 2026-09-28 01:00 UTCGHSA enrichmentGHSA-jrf5-f6jp-7vxc · GitHub severity HIGH · GitHub CVSS 6.5
  27. 2026-09-27 13:48 UTCEPSS rescoreEPSS 0.5% · 37th percentile
  28. 2026-09-27 02:03 UTCEG score recompute
  29. 2026-09-27 02:03 UTCGHSA enrichmentGHSA-jrf5-f6jp-7vxc · GitHub severity HIGH · GitHub CVSS 6.5
  30. 2026-09-26 15:59 UTCEPSS rescoreEPSS 0.5% · 37th percentile
  31. 2026-09-26 02:27 UTCEG score recompute
  32. 2026-09-26 02:27 UTCGHSA enrichmentGHSA-jrf5-f6jp-7vxc · GitHub severity HIGH · GitHub CVSS 6.5
  33. 2026-09-24 20:35 UTCEG score recompute
  34. 2026-09-24 20:35 UTCGHSA enrichmentGHSA-jrf5-f6jp-7vxc · GitHub severity HIGH · GitHub CVSS 6.5
  35. 2026-09-24 14:04 UTCEPSS rescoreEPSS 0.5% · 37th percentile
  36. 2026-09-24 00:37 UTCEG score recompute
  37. 2026-09-24 00:37 UTCGHSA enrichmentGHSA-jrf5-f6jp-7vxc · GitHub severity HIGH · GitHub CVSS 6.5
  38. 2026-09-23 17:54 UTCEPSS rescoreEPSS 0.3% · 24th percentile
  39. 2026-09-23 01:41 UTCEG score recompute
  40. 2026-09-23 01:41 UTCGHSA enrichmentGHSA-jrf5-f6jp-7vxc · GitHub severity HIGH · GitHub CVSS 6.5
  41. 2026-09-22 16:01 UTCEPSS rescoreEPSS 0.3% · 24th percentile
  42. 2026-09-22 02:43 UTCEG score recompute
  43. 2026-09-22 02:43 UTCGHSA enrichmentGHSA-jrf5-f6jp-7vxc · GitHub severity HIGH · GitHub CVSS 6.5
  44. 2026-09-21 21:09 UTCEPSS rescoreEPSS 0.3% · 24th percentile
  45. 2026-09-21 03:46 UTCEG score recompute
  46. 2026-09-21 03:46 UTCGHSA enrichmentGHSA-jrf5-f6jp-7vxc · GitHub severity HIGH · GitHub CVSS 6.5
  47. 2026-09-20 20:16 UTCEPSS rescoreEPSS 0.3% · 24th percentile
  48. 2026-09-20 04:49 UTCEG score recompute
  49. 2026-09-20 04:49 UTCGHSA enrichmentGHSA-jrf5-f6jp-7vxc · GitHub severity HIGH · GitHub CVSS 6.5
  50. 2026-09-19 05:52 UTCEG score recompute
  51. 2026-09-19 05:52 UTCGHSA enrichmentGHSA-jrf5-f6jp-7vxc · GitHub severity HIGH · GitHub CVSS 6.5
  52. 2026-09-18 19:28 UTCEPSS rescoreEPSS 0.3% · 24th percentile
  53. 2026-09-18 06:55 UTCEG score recompute
  54. 2026-09-18 06:55 UTCGHSA enrichmentGHSA-jrf5-f6jp-7vxc · GitHub severity HIGH · GitHub CVSS 6.5
  55. 2026-09-17 19:32 UTCEPSS rescoreEPSS 0.3% · 23rd percentile
  56. 2026-09-17 07:58 UTCEG score recompute
  57. 2026-09-17 07:58 UTCGHSA enrichmentGHSA-jrf5-f6jp-7vxc · GitHub severity HIGH · GitHub CVSS 6.5
  58. 2026-09-16 14:09 UTCEPSS rescoreEPSS 0.3% · 23rd percentile
  59. 2026-09-16 09:01 UTCEG score recompute
  60. 2026-09-16 09:01 UTCGHSA enrichmentGHSA-jrf5-f6jp-7vxc · GitHub severity HIGH · GitHub CVSS 6.5
  61. 2026-09-16 05:15 UTCEPSS rescoreEPSS 0.3% · 23rd percentile
  62. 2026-09-15 08:59 UTCEG score recompute
  63. 2026-09-15 08:59 UTCGHSA enrichmentGHSA-jrf5-f6jp-7vxc · GitHub severity HIGH · GitHub CVSS 6.5
  64. 2026-09-15 03:12 UTCEPSS rescoreEPSS 0.3% · 23rd percentile
  65. 2026-09-14 10:02 UTCEG score recompute
  66. 2026-09-14 10:02 UTCGHSA enrichmentGHSA-jrf5-f6jp-7vxc · GitHub severity HIGH · GitHub CVSS 6.5
  67. 2026-09-13 11:05 UTCEG score recompute
  68. 2026-09-13 11:05 UTCGHSA enrichmentGHSA-jrf5-f6jp-7vxc · GitHub severity HIGH · GitHub CVSS 6.5
  69. 2026-09-12 12:08 UTCEG score recompute
  70. 2026-09-12 12:08 UTCGHSA enrichmentGHSA-jrf5-f6jp-7vxc · GitHub severity HIGH · GitHub CVSS 6.5
  71. 2026-09-11 14:53 UTCEPSS rescoreEPSS 0.3% · 23rd percentile
  72. 2026-09-11 13:11 UTCEG score recompute
  73. 2026-09-11 13:11 UTCGHSA enrichmentGHSA-jrf5-f6jp-7vxc · GitHub severity HIGH · GitHub CVSS 6.5

Related CVEs(same CWE)

Frequently asked(5)

What is CVE-2026-85670?
CVE-2026-85670 is a medium-severity vulnerability; its CVE record was published on September 4, 2026. 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::fromfile/fromstr, the builder sizes a scratch buffer to the longest vocabulary key, then writes each concatenated merge…
When was CVE-2026-85670 made public?
CVE-2026-85670 was made public on June 10, 2026: that is the date its CVE record's CNA, VulnCheck, gives for when the vulnerability was made public. The CVE record itself was published on September 4, 2026. EchelonGraph's copy of the record carries a last-modified date of September 24, 2026; the record at its source may have been updated since.
Is CVE-2026-85670 actively exploited?
CVE-2026-85670 is not currently on CISA's Known Exploited Vulnerabilities catalog. FIRST EPSS estimates a 0.5% probability of exploitation in the next 30 days (39th percentile of EPSS-scored CVEs).
What is the CVSS score of CVE-2026-85670?
CVE-2026-85670 has a CVSS base score of 6.5 (a secondary CVSS source that NVD displays; NVD's own analysis pending).
How do I remediate CVE-2026-85670?
No fix for CVE-2026-85670 is confirmed yet. Until one is published, restrict network exposure of the affected system or apply the vendor's mitigation — for example, keep it off the internet or limit it to trusted networks — and watch the vendor's advisory for the fix. The CVE record's CNA (VulnCheck) lists huggingface tokenizers versions up to and including 0.23.2 as affected. No release containing a fix is confirmed on record. The vendor advisories EchelonGraph has for CVE-2026-85670 are linked in the Vendor Advisories panel on this page.

Dependency Blast Radius

Explore the affected products and dependency analysis for CVE-2026-85670

Explore →

Is Your Infrastructure Affected by CVE-2026-85670?

EchelonGraph automatically scans your cloud infrastructure and maps CVE exposure using blast radius analysis.

Book a Demo →Browse All CVEs