AI Security Threat Feed

Latest CVEs affecting AI/ML systems — LLM frameworks, ML libraries, AI agents, vector databases, and inference servers. Vulnerabilities are tracked from NVD, GitHub Advisory, CISA KEV, MITRE ATLAS, and enriched with CVSS, EPSS, exploitation confidence, AI-component classification, and compliance mappings to ISO 42001, EU AI Act, NIST AI RMF, and OWASP LLM Top 10. Updated continuously as new CVEs are published.

Each CVE is enriched with
  • CVSS severity
  • EPSS exploit probability
  • Exploitation confidence
  • AI-component classification
  • Compliance mappings
3,589

AI/ML CVEs Tracked

481

Critical

94

New This Week

24

In CISA KEV

Latest AI Security Threats

Showing 20 of 1586 results — High severity
HIGH

vLLM Mooncake connector through 0.29.0 fails to properly manage GPU KV cache block ownership when concurrent child requests share a single transfer ID in...

CVE-2026-94627
7.5
CWE-401
HIGH

vLLM through 0.29.0 fails to validate the tp_size parameter in kv_transfer_params on OpenAI-compatible completion endpoints, allowing attackers to allocate...

CVE-2026-94626
7.5
CWE-789
HIGH

vLLM through 0.29.0 contains a denial of service vulnerability in P2P KV offloading when OffloadingConnector is configured with TieringOffloadingSpec and a...

CVE-2026-94624
7.5
CWE-770
HIGH

vLLM through 0.29.0 contains a denial of service vulnerability in the NIXL connector's prefix caching implementation that fails to properly validate block...

CVE-2026-94623
7.5
CWE-617
HIGH

vLLM versions through 0.29.0 contain a denial of service vulnerability in the NIXL connector's metadata handling for prefill/decode disaggregated deployments....

CVE-2026-94622
7.5
CWE-248
HIGH

Joplin is an open source note-taking and to-do application that organises notes and lists into notebooks. Prior to 3.7.7, Joplin Server's GET...

CVE-2026-59814
7.6
CWE-79
HIGH

Joplin is an open source note-taking and to-do application that organises notes and lists into notebooks. Prior to 3.7.2, Joplin Server's UserModel.ssoLogin()...

CVE-2026-55210
7.4
CWE-290
HIGH

Joplin is an open source note-taking and to-do application that organises notes and lists into notebooks. Prior to 3.6.15 and 3.7.2,...

CVE-2026-55105
7.7
CWE-79
HIGH

Joplin is an open source note-taking and to-do application that organises notes and lists into notebooks. Prior to 3.6.15 and 3.7.2, Joplin accepts...

CVE-2026-49453
7.0
CWE-20
HIGH

Joplin is an open source note-taking and to-do application that organises notes and lists into notebooks. Prior to 3.7.2, Joplin Desktop for Windows omits...

CVE-2026-49450
7.1
CWE-345
HIGH

Unrestricted Upload of File with Dangerous Type in core/modules/media.php in CuteNews v.2.1.2 allows remote authenticated users with access to the Media...

CVE-2026-36467
7.2
HIGH

Mistral Vibe before 2.25.5 contains a remote code execution vulnerability in the worktree creation process that executes git hooks before trust validation....

CVE-2026-93993
8.8
EPSS 0.6%
mistral-vibe 492
HIGH

SiYuan through 3.8.4 fails to escape heading style attributes when rendering outline and bookmark dock HTML, allowing stored cross-site scripting. Attackers...

CVE-2026-93923
8.8
EPSS 0.4%
CWE-79
HIGH

SiYuan through 3.8.4 renders notebook names as raw HTML in the Daily Note picker dialog without escaping, allowing stored cross-site scripting in the Electron...

CVE-2026-93922
8.8
EPSS 0.5%
CWE-79
HIGH EXPLOIT AVAIL

ToolHive: containerized MCP servers can reach host services via host.docker.internal, enabling lateral movement

CVE-2026-58197
8.8
EPSS 0.4%
github.com/stacklok/toolhive Patch: 0.30.1 CWE-284 1.8K
HIGH

LMDeploy has an SSRF bypass

GHSA-39wr-7q6h-cf68
7.5
lmdeploy Patch: 0.15.0 CWE-436 1
HIGH

LMDeploy is a toolkit for compressing, deploying, and serving large language models. Versions 012.1 through 0.12.2 contain a code injection vulnerability in...

CVE-2026-33625
8.8
EPSS 0.2%
lmdeploy Patch: 0.12.3 CWE-400 1
HIGH EXPLOIT AVAIL

vLLM versions before 0.28.0 fail to validate the lower bound of token IDs in the /v1/embeddings and /pooling endpoints, allowing unauthenticated attackers to...

CVE-2026-93592
7.5
EPSS 0.4%
CWE-129
HIGH EXPLOIT AVAIL

SiYuan versions before 3.8.3 contain an SQL injection vulnerability in the graph.go query2Stmt function where tag values are concatenated raw into SQL string...

CVE-2026-93591
7.6
EPSS 0.3%
CWE-89
HIGH

vLLM through 0.29.0 fails to properly clean up decode-side metadata for rejected inference requests in prefill/decode disaggregated deployments. Remote...

CVE-2026-93436
7.5
EPSS 0.5%
CWE-401

Frequently asked questions

What is an AI security threat feed?

An AI security threat feed is a continuously updated stream of vulnerabilities (CVEs) affecting AI and machine-learning systems — LLM frameworks, ML libraries, AI agents, vector databases, and inference servers — filtered out of the broader CVE firehose and enriched for relevance.

Which sources are the AI CVEs tracked from?

CVEs are tracked from NVD, GitHub Advisory, CISA KEV, and MITRE ATLAS, then enriched with CVSS, EPSS, exploitation confidence, AI-component classification, and compliance mappings.

What AI systems do these vulnerabilities affect?

Coverage spans LLM frameworks, ML libraries, AI agents, vector databases, and inference servers — the components most security teams now run in production.

How often is the AI threat feed updated?

The feed updates continuously as new CVEs are published and enriched, so the most recent AI/ML vulnerabilities appear at the top.

Is the AI security feed free?

Yes — the public feed is free to browse. A Pro subscription adds breaking alerts, MITRE ATLAS mappings, compliance reports (ISO 42001, EU AI Act), and full CISO analysis.

Need deeper analysis?

Get ATLAS technique mappings, compliance reports (ISO 42001, EU AI Act), breaking alerts, and full CISO analysis with a Pro subscription.

Start 14-Day Free Trial