CVE-2026-93840

A flaw was found in vLLM. This vulnerability arises from incorrect validation of allowedtokenids against the tokenizer length instead of the model's output logits width. A remote attacker can exploit this by supplying token IDs that exceed the output vocabulary, which bypasses validation. This ca...

CVE-2026-93989

A flaw was found in vLLM. A remote attacker can exploit a vulnerability by providing malformed input related to 'bad words' token processing. This can lead to out-of-bounds memory access, corrupting the internal data logits memory used for generating responses for other concurrent requests. As a...

Improper Validation of Array Index

Overview vllm is an A high-throughput and memory-efficient inference and serving engine for LLMs Affected versions of this package are vulnerable to Improper Validation of Array Index via SamplingParams.updatefromtokenizer, which fails to validate badwords token indices against the model's...

EUVD-2026-83617

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

CVE-2026-93989

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

CVE-2026-93989 vLLM through 0.29.0 Cross-Request Logits Corruption via bad_words

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

CVE-2026-93989 vLLM through 0.29.0 Cross-Request Logits Corruption via bad_words

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

CVE-2026-93989

The vLLM inference and serving engine (through version 0.29.0 ) is vulnerable to an Improper Validation of Array Index in the SamplingParams.update_from_tokenizer() function. A remote authenticated attacker can supply out-of-bounds bad_words token indices that fail to be validated against the mod...

CVE-2026-93989 vLLM through 0.29.0 Cross-Request Logits Corruption via bad_words

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

CVE-2026-93989: Improper Validation of Array Index

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

Improper Validation of Array Index

Overview vllm is an A high-throughput and memory-efficient inference and serving engine for LLMs Affected versions of this package are vulnerable to Improper Validation of Array Index via the stoptokenids and allowedtokenids parameters in vllm/samplingparams.py, which are not validated against th...

EUVD-2026-83284

vLLM before 0.29.0 validates allowedtokenids against tokenizer length instead of model output logits width in SamplingParams.validateallowedtokenids. Attackers can supply token IDs above the output vocabulary that pass validation, causing LogitBiasState to corrupt GPU logits state and allow...

CVE-2026-93840 vLLM before 0.29.0 Cross-Request Logits Corruption via allowed_token_ids

vLLM before 0.29.0 validates allowedtokenids against tokenizer length instead of model output logits width in SamplingParams.validateallowedtokenids. Attackers can supply token IDs above the output vocabulary that pass validation, causing LogitBiasState to corrupt GPU logits state and allow...

CVE-2026-93840

vLLM before 0.29.0 is vulnerable to cross-request logits corruption via the allowed_token_ids parameter. The root cause is in SamplingParams._validate_allowed_token_ids() , which validates token IDs against the tokenizer length rather than the model output logits width. An attacker can supply tok...

CVE-2026-93840 vLLM before 0.29.0 Cross-Request Logits Corruption via allowed_token_ids

vLLM before 0.29.0 validates allowedtokenids against tokenizer length instead of model output logits width in SamplingParams.validateallowedtokenids. Attackers can supply token IDs above the output vocabulary that pass validation, causing LogitBiasState to corrupt GPU logits state and allow...

CVE-2026-93840 vLLM before 0.29.0 Cross-Request Logits Corruption via allowed_token_ids

vLLM before 0.29.0 validates allowedtokenids against tokenizer length instead of model output logits width in SamplingParams.validateallowedtokenids. Attackers can supply token IDs above the output vocabulary that pass validation, causing LogitBiasState to corrupt GPU logits state and allow...

CVE-2026-93840: Improper Validation of Array Index

vLLM before 0.29.0 validates allowedtokenids against tokenizer length instead of model output logits width in SamplingParams.validateallowedtokenids. Attackers can supply token IDs above the output vocabulary that pass validation, causing LogitBiasState to corrupt GPU logits state and allow...

RobPI: Robust Private Inference against Malicious Client

The increased deployment of machine learning inference in various applications has sparked privacy concerns. In response, private inference PI protocols have been created to allow parties to perform inference without revealing their sensitive data. Despite recent advances in the efficiency of PI,...

LLM Jailbreak Detection for (Almost) Free!

Large language models LLMs enhance security through alignment when widely used, but remain susceptible to jailbreak attacks capable of producing inappropriate content. Jailbreak detection methods show promise in mitigating jailbreak attacks through the assistance of other models or multiple model...

Busting the Paper Ballot: Voting Meets Adversarial Machine Learning

We show the security risk associated with using machine learning classifiers in United States election tabulators. The central classification task in election tabulation is deciding whether a mark does or does not appear on a bubble associated to an alternative in a contest on the ballot. Barrett...