Deck for 2026-11-14 (UTC)11 briefs
ThreatDeck — 2026-11-14
11 cards today · 2 actively exploited or must-patch · 6 affecting AI tooling.
- 01
Act Now Critical KEV CVE-2026-123039 AI-assisted
CISA KEV: Check Point Gaia OS DynamicCLI local privilege escalation to root (CVE-2026-123039)
CISA added CVE-2026-123039 to its Known Exploited Vulnerabilities catalog on 2026-11-13. Check Point Firewall, Multi-Domain Security Management Server, Scalable Platforms, Security Gateway, and Security Management Server are affected. The Gaia OS DynamicCLI daemon runs as root and listens on a world-writable local socket without authenticating callers. Federal agencies must apply mitigations by 2026-11-16.
Why it matters. An authenticated local user on an affected Check Point appliance can run commands as root, gaining full control of a security gateway or management server. Compromise of these devices can undermine network perimeter controls and policy management.
Do this. Apply the vendor mitigations described in Check Point advisory sk1000362 on all affected gateways and management servers.
2026-11-13 · score 62 · CISA Known Exploited Vulnerabilities
- 02
Act Now Critical KEV CVE-2026-114713 AI-assisted
CISA adds UnrealIRCd delimiter injection flaw CVE-2026-114713 to KEV catalog
CISA added CVE-2026-114713 in UnrealIRCd to its Known Exploited Vulnerabilities catalog on 2026-11-13. The flaw is an improper neutralization of delimiters issue. It lets any user on a network with two or more linked servers inject server-to-server commands that run with server authority. The due date for required action is 2026-11-16.
Why it matters. Any user on an affected linked-server IRC network could run commands with server authority, which can compromise network integrity. The short three-day remediation deadline signals active exploitation risk for exposed UnrealIRCd deployments.
Do this. Identify all UnrealIRCd instances, especially networks with two or more linked servers.
2026-11-13 · score 61 · CISA Known Exploited Vulnerabilities
- 03
AI Stack Alert Critical CVE-2026-123965 AI-assisted
Jet Admin Cloud Platform prompt injection (CVE-2026-123965) exposes .env files and credentials
CVE-2026-123965 is a prompt injection vulnerability in all versions of Jet Admin Cloud Platform. A remote attacker with a free account can expose sensitive .env configuration files and exfiltrate critical credentials. The source says it was mitigated via a sandbox component around October 29, 2026, and gives a CVSS base score of 9.3.
Why it matters. Any organization using Jet Admin Cloud Platform could have had credentials in .env files exposed to anyone able to create a free account. Stolen credentials could give an attacker access to connected systems and data.
Do this. Confirm with Jet Admin that the sandbox mitigation applies to your tenant and is active.
2026-11-13 · score 39 · NVD (AI keyword search)
- 04
AI Stack Alert High TD-2026-0094 AI-assisted
FBI flash alert: Artex open-source AI tool automates cyberattacks at machine speed
The FBI issued a flash alert on Artex, an open-source AI platform that can use any large language model to automate multiple phases of intrusions. The FBI says some intrusions reached privileged access in under three hours. Censys research cited by the FBI found over 1,600 publicly exposed Artex console instances.
Why it matters. Organizations with internet-facing or third-party-connected systems face faster, higher-volume intrusions that can outpace manual defenses. Less-monitored partner portals, legacy services, and vendor integrations are named as likely targets.
Do this. Inventory internet-exposed services and third-party integrations, prioritizing partner portals, legacy services, and employee applications for review.
2026-11-13 · score 36 · The Record
- 05
AI Stack Alert High CVE-2026-68905 AI-assisted
CVE-2026-68905: Hugo Blog AI assistant proxy lacks authentication, enabling unauthenticated API key abuse
Hugo Blog commit 744ca15 and prior expose three AI assistant proxy endpoints without authentication checks. When deployed with the documented Nginx reverse-proxy setup, unauthenticated remote attackers can forward arbitrary requests to upstream LLM, embedding, and rerank APIs using the operator's server-side API key. No public patch is available at publication.
Why it matters. Exposed operators face unlimited quota consumption, cost exhaustion, and possible service unavailability. An attacker gains use of the operator's paid AI API credentials without any authorization.
Do this. Restrict access to /api/blog-assistant, /api/blog-search-embedding, and /api/blog-search-rerank at the Nginx layer, requiring authentication until a fix is available.
2026-11-13 · score 36 · NVD (AI keyword search)
- 06
AI Stack Alert High CVE-2026-126399 AI-assisted
CVE-2026-126399: vLLM 0.19.0 to before 0.31.0 resource exhaustion via video num_frames bypass
vLLM from 0.19.0 before 0.31.0 contains a resource exhaustion vulnerability. Inference clients can bypass the server frame limit by setting media_io_kwargs.video.num_frames in chat completion requests. This forces VideoMediaIO to decode all frames and stack them, causing memory exhaustion.
Why it matters. Any exposed vLLM inference endpoint accepting chat completion requests with video input can be driven to memory exhaustion, causing denial of service for other users of the server. The CVSS base score is 7.1.
Do this. Upgrade vLLM to version 0.31.0 or later.
2026-11-13 · score 36 · NVD (AI keyword search)
- 07
AI Stack Alert High CVE-2026-126398 AI-assisted
CVE-2026-126398: vLLM 0.28.0 to before 0.30.0 reachable assertion lets remote attackers crash engine
vLLM from 0.28.0 before 0.30.0 contains a reachable assertion in GPUModelRunner._init_xdrope_positions. A remote attacker can crash the engine by sending a /v1/completions request containing only prompt_embeds to XD-RoPE models run with --enable-prompt-embeds. The crash terminates the EngineCore and API server for all clients. The CVSS base score is 8.2.
Why it matters. Any exposed vLLM deployment serving XD-RoPE models with prompt embeddings enabled can be taken offline by one unauthenticated request. This affects availability for all users of the service.
Do this. Upgrade vLLM to version 0.30.0 or later.
2026-11-13 · score 36 · NVD (AI keyword search)
- 08
AI Stack Alert High CVE-2026-126397 AI-assisted
vLLM 0.10.1 to 0.31.0: request-controlled max_patches enables CPU denial of service on Cohere2 Vision
vLLM versions from 0.10.1 before 0.31.0 pass a request-controlled mm_processor_kwargs.max_patches value to the Hugging Face processor for Cohere2 Vision models. The source reports no server-side upper bound on this value. Remote clients can send small image requests with large, varying max_patches values to bypass the LRU cache and exhaust preprocessing workers.
Why it matters. Any remote client that can reach an affected vLLM endpoint can degrade service for other users. The source assigns a CVSS base score of 8.7.
Do this. Upgrade vLLM to version 0.31.0 or later.
2026-11-13 · score 36 · NVD (AI keyword search)
- 09
New Technique High TD-2026-0091 AI-assisted
Cloudflare reports surge in LLM pentest orchestrators like ARTEX and forks scanning customers
Cloudflare reports that over 50 LLM penetration testing framework families attempted to scan its customers in three days. ARTEX was taken down by its creator on October 8, but the ecosystem has since produced many forks and hundreds of similar frameworks. Cloudflare says some attacks have chained stolen customer credentials for lateral movement.
Why it matters. Autonomous orchestrators let a single operator target many organizations at once and iterate through attack stages quickly. Organizations with static, non-expiring CI/CD tokens or weak bot and rate controls face greater exposure to credential-chained intrusions.
Do this. Rotate and replace static, non-expiring API and CI/CD tokens, and enforce IP restrictions and rate limits on account-level tokens.
2026-11-13 · score 33 · Cloudflare Blog: Security
- 10
New Technique Medium TD-2026-0085 AI-assisted
Post-training supply chain attacks can backdoor coding agents via poisoned datasets or model checkpoints
Microsoft Defender Research showed that a small number of crafted examples in a fine-tuning dataset can teach a coding agent model to misuse command-execution tools when a routine request appears. In a separate study, the Microsoft AI Red Team found that backdoored open-weight checkpoints kept high activation rates on unseen repositories, reaching 84.1% to 94.8%. Microsoft recommends checkpoint provenance, dataset integrity checks, and strict tool permissions.
Why it matters. Organizations that fine-tune models or run third-party open-weight checkpoints in coding agents may execute attacker-chosen commands with the agent's credentials and file access. The model can behave normally on standard benchmarks, so routine evaluation may not reveal the backdoor.
Do this. Pin reviewed training datasets by cryptographic hash and investigate unexpected changes.
2026-11-12 · score 25 · Microsoft Security Blog
- 11
New Technique Medium TD-2026-0086 AI-assisted
S1-Evo: self-evolving attacks manipulate typed System-1 decision model outputs
The paper proposes S1-Evo, a framework that perturbs untrusted fields in inputs to typed System-1 decision models so that the models' typed outputs and execution branches change. Across four benchmarks and three S1 models, the authors report 71.99% mean attack success rate and 57.47% high-confidence ASR on jev-1.14, and up to 24.00% end-to-end agent violations.
Why it matters. LLM agents that use typed S1 models for routing, moderation, or authorization can have those decisions altered by untrusted context. An attacker who controls such context could change which execution branch an agent takes.
Do this. Inventory LLM agent components that use typed classifier or scoring models for routing, moderation, or authorization decisions.
2026-11-12 · score 23 · arXiv (AI security papers)