CVE-2026-93989 PUBLISHED

vLLM through 0.29.0 Cross-Request Logits Corruption via bad_words

Assigner: VulnCheck
Reserved: 19.09.2026 Published: 19.09.2026 Updated: 19.09.2026

vLLM through 0.29.0 fails to properly validate bad_words token indices against the model's generation output width in SamplingParams.update_from_tokenizer(). Attackers can supply out-of-bounds token indices that corrupt logits memory of concurrent requests, causing different in-flight HTTP requests to return incorrect tokens.

CVSS Vector: CVSS:4.0/AV:N/AC:H/AT:P/PR:L/UI:N/VC:N/VI:L/VA:N/SC:N/SI:N/SA:N
CVSS Score: 2.3

Exploitability Metrics		Vulnerable System Impact Metrics		Subsequent System Impact Metrics
Attack Vector	Network	Confidentiality	None	Confidentiality	None
Attack Complexity	High	Integrity	Low	Integrity	None
Attack Requirements	Present	Availability	None	Availability	None
Privileges Required	Low
User Interaction	None

CVSS 4.0

CVSS Vector: CVSS:3.1/AV:N/AC:H/PR:L/UI:N/S:U/C:N/I:L/A:N
CVSS Score: 3.1

Attack Vector	Network	Scope	Unchanged
Attack Complexity	High	Confidentiality Impact	None
Privileges Required	Low	Integrity Impact	Low
User Interaction	None	Availability Impact	None

CVSS 3.1

Product Status

Vendor	vllm-project
Product	vllm
Versions	Default: unaffected affected from 0 to 0.29.0 (incl.)

CVE-2026-93989 PUBLISHED

vLLM through 0.29.0 Cross-Request Logits Corruption via bad_words

Metrics

Product Status

Credits

References

Problem Types