ai_security▌

2026-11-14

9 item(s) verified and swept on 2026-11-14, each citing a primary source.

NOTABLECVETOOL

vLLM publishes six GHSAs: multimodal-preprocessing DoS batch, engine crash, and a heap-address leak

Six vLLM advisories with CVEs (CVE-2026-126394 through 126399) were published in one NVD batch: request-controlled mm_processor_kwargs.max_patches forces quadratic CPU work on Cohere2 Vision preprocessing (126397, high), num_frames and min_pixels values bypass media-processing budgets to OOM or exhaust the API server (126394, 126396, 126399), embeddings-only /v1/completions requests trip an assertion that kills the engine on XD-RoPE models with --enable-prompt-embeds (126398, high, fixed in 0.30.0), and a lowercase-only sanitize_message() regex leaks live server heap addresses through uppercase PIL image addresses in crafted PNGs, an ASLR-bypass building block (126395). Five are fixed in 0.31.0.

NOTABLECVETOOL

Apache APISIX MCP endpoint could proxy calls to routes its own OpenAPI document never declared (CVE-2026-124726)

A relative-path-traversal, unintended-proxy flaw lets a caller of the APISIX MCP endpoint invoke endpoints the OpenAPI document never declared, reusing the declared tool's HTTP method and the credentials the Route adds; affects APISIX 3.19.0 before 3.20.0, fixed in 3.20.0.

NOTABLETOOL

NLTK publishes a batch of security advisories for its parsers and corpus readers

The NLTK repository advisory list now carries a batch of reviewed advisories: unbounded CPU and quadratic-time denial of service in SteppingRecursiveDescentParser.parse(), the CCG lexicon parser, ConllCorpusReader, TreebankWordTokenizer and LegalitySyllableTokenizer on crafted grammars or corpora, plus path traversal in save_maxent_params/load_maxent_params; several carry CVEs (e.g. CVE-2026-126303), all published to the global advisory database on Nov 13.

NOTABLEPOLICY

Project Zero replaces its 90+30 policy with 'Full 90' stepping to 60 then 30 days, citing LLM-accelerated exploitation

Natalie Silvanovich: with LLMs raising discovery rates to several vulnerabilities a week, driving patch-to-exploitation time toward zero, and making vendor negotiation unmanageable, the 90+30 timeline is 'clearly insufficient'. Project Zero now discloses 90 days after report ('Full 90'), moving to 'Full 60' on Feb 2, 2027 and 'Full 30' on Jan 3, 2028, removing grace-period and variant special-cases; actively exploited bugs follow Google's application-security policy.

NOTABLETOOL

Claude Code v2.1.325 fixes a security issue in how zstd-compressed responses are decoded

The v2.1.325 release notes (Nov 13 20:40 UTC) carry one security line: 'Fixed a security issue in how zstd-compressed responses are decoded'. No CVE, advisory, or technical detail was published. Added under the spec section 2 changelog rule, same pattern as the v2.1.319 and v2.1.321 release-note items.

NOTABLEPROMPT-INJECTION

arXiv 2611.14548 SideTaskBench: third-party references steer coding agents into producing and delivering work the user never requested

SideTaskBench pairs 25 Python repairs with 190 self-contained side tasks to measure a failure repair benchmarks miss: third-party technical references inducing an agent to complete and deliver independent side-task artifacts during an authorized repair, scored across a four-level delivery model alongside repair success.

NOTABLESUPPLY-CHAIN

arXiv 2611.14978: a disguised LoRA adapter passes audit on the unquantized base, then hijacks decisions once deployment quantizes it

An attacker-supplied adapter for decision tasks such as content moderation and function selection preserves benign behavior on the unquantized trusted base — passing audit — but hijacks decisions after the same base is quantized for deployment, without modifying the base and without knowing the victim's quantization configuration.

NOTABLEPOLICY

arXiv 2611.16122 Intent-Aware Security makes the trusted agent intent a first-class security object

Actions can contain no attack pattern and stay within granted permissions yet still violate an agent's role, organizational constraints, or deployment purpose. The paper proposes treating the trusted agent intent as a first-class security object — intent fidelity as a control that attack-pattern, output, and behavior detection do not provide.

NOTABLERED-TEAM

arXiv 2611.16158 PolicySniff: a co-resident GPU tenant infers which approval policy an agent enforces from kernel timing

Operators hide agent approval settings by restricting prompt and response access, but PolicySniff shows a co-resident GPU process can infer the deployed setting: profile both candidate settings offline, then time only its own kernels while the agent processes tasks, recovering the agent's approval-policy state (e.g. ask-before-consequential-tool-call) through shared-GPU execution.

← 2026-11-16 · 2026-11-13 → · all days