Raw vector
CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:HSummary
CVE-2025-29770 is a medium-severity Allocation of Resources Without Limits or Throttling (CWE-770) vulnerability in Vllm Vllm. Its CVSS base score is 6.5 (Medium).
Operationally, exploitation aligns with the MITRE ATT&CK technique Network Denial of Service (T1498); ranked at the 36th percentile by exploit likelihood (below the median); it is not currently listed in the CISA KEV catalog.
This vulnerability is AI-related — categorised as NLP and Transformers; in the LLM/Generative AI Risks risk domain.
The strongest mitigations our analysis identified map to AC-10 (Concurrent Session Control) and SC-6 (Resource Availability) — see the control section below for these in your framework.
EU & UK References
- 🇪🇺 ENISA EUVD: EUVD-2025-6726
Vulnerability Data
vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs. The outlines library is one of the backends used by vLLM to support structured output (a.k.a. guided decoding). Outlines provides an optional cache for its compiled grammars on…
more
the local filesystem. This cache has been on by default in vLLM. Outlines is also available by default through the OpenAI compatible API server. The affected code in vLLM is vllm/model_executor/guided_decoding/outlines_logits_processors.py, which unconditionally uses the cache from outlines. A malicious user can send a stream of very short decoding requests with unique schemas, resulting in an addition to the cache for each request. This can result in a Denial of Service if the filesystem runs out of space. Note that even if vLLM was configured to use a different backend by default, it is still possible to choose outlines on a per-request basis using the guided_decoding_backend key of the extra_body field of the request. This issue applies only to the V0 engine and is fixed in 0.8.0.
- CWE(s)
AI Security AnalysisAI
- AI Category
- NLP and Transformers
- Risk Domain
- LLM/Generative AI Risks
- OWASP Top 10 for LLMs 2025
- None mapped
- Classification Reason
- Matched keywords: llms, openai, vllm
Related Threats
MITRE ATT&CK Enterprise Techniques
CVEs Like This One
Affected Assets
Mitigating Controls
Control response
—
—
- 5 hardening rules · 3 OS baselines
V15.4.4
Mitigating Controls (NIST 800-53 r5) AI
Directly enforces a hard limit on concurrent sessions, structurally preventing unbounded resource allocation.
Requires explicit allocation of resources by priority or quota, directly stopping unlimited allocation.
Imposes a limit on consecutive invalid attempts, preventing one specific class of unbounded resource consumption.
Reduces impact of resulting DoS events without preventing the underlying lack of allocation limits.
Mitigating Controls (NIST CSF 2.0) AI
Derived directly from the weakness types (CWEs) cited in the NVD entry via our AI-authored CWE→CSF cross-walk (authority under review) — links open the control.
Monitoring capacity and taking action to maintain availability directly reduces unchecked resource allocation.
Mitigating Controls (ISO/IEC 27001:2022 Annex A) AI
Derived directly from the weakness types (CWEs) cited in the NVD entry via our AI-authored CWE→ISO cross-walk (authority under review) — links open the control.
Baseline comparison of CPU, memory and bandwidth usage helps surface uncontrolled resource allocations before they cause service degradation.
Capacity projections and elasticity measures ensure that allocation requests are bounded and can be throttled, reducing the window in which an attacker can force unbounded resource reservations.
Defining retention periods and deletion schedules for backup copies prevents indefinite accumulation of data on storage media without corresponding resource-management controls.
Architectural redundancy and automatic failover limit the impact of an attacker who forces excessive allocations, because spare capacity can absorb the load until the primary instance recovers.
Documented incident response procedures that include activation of continuity plans and controlled recovery help ensure that resource consumption triggered by an incident is bounded and managed rather than left unbounded.
Mandating tested continuity procedures that preserve or replace resource-limiting controls prevents an attacker from exploiting the absence of throttling mechanisms during an outage.
Hardening callouts derived
Configuration rules from DISA STIG baselines that bear on weaknesses of the type cited by this CVE. Each rule is shown with the relationship its mapping actually records, against the CWE it was authored against. Derived via CVE→CWE over `controls_xwalks` (authoritative rows only; rows rated `none` are excluded).
Oracle Linux 8 (2 rules)
- V-248552 OL 8 must be configured so that all network connections associated with SSH traffic terminate after becoming unresponsive. prevents CWE-770
- V-248553 OL 8 must be configured so that all network connections associated with SSH traffic are terminated after 10 minutes of becoming unresponsive. prevents CWE-770
Oracle Linux 9 (2 rules)
- V-271710 OL 9 must be configured so that all network connections associated with SSH traffic are terminated after 10 minutes of becoming unresponsive. prevents CWE-770
- V-271709 OL 9 must be configured so that all network connections associated with SSH traffic terminate after becoming unresponsive. prevents CWE-770
RHEL 8 (1 rule)
- V-230244 RHEL 8 must be configured so that all network connections associated with SSH traffic terminate after becoming unresponsive. prevents CWE-770