{"date": "2026-10-29", "generated_at": "2026-10-29T03:03:25.608800+00:00", "headline": "vLLM patches land just in time; cloud account hygiene takes the rest of the day", "count": 9, "full_edition": "/content/cybersecurity/2026-10-29.json", "top10": [{"priority_rank": 1, "cve_id": "CVE-2026-115915", "name": "vLLM: Untrusted EPS inputs can trigger denial of service or Ghostscript RCE", "vendor": "", "product": "vllm", "cwe": ["CWE-183", "CWE-20"], "epss": 0.00631, "priority_reason": "AI blast radius (vllm); no exploitation reported; patch available (0.32.0)", "narrative": "This issue in vLLM 0.30.0 allows untrusted EPS image inputs to trigger a denial of service or remote code execution via Ghostscript. The vulnerability stems from `ImageMediaIO` passing attacker-controlled bytes to Pillow without restricting formats, meaning a malicious EPS labeled as PNG can reach the host’s Ghostscript executable. It ranks high because it touches the AI layer with a blast radius of 3, and while no exploitation is reported yet, the EPSS is 0.00631. In a Kubernetes estate, this affects any pod running vLLM with multimodal support, where an attacker could send a crafted image to a public endpoint. The concrete action is to upgrade to vLLM 0.32.0, which is available now. This is a straightforward patch that removes the risk of sandbox escape via Ghostscript, so it is worth patching this week to keep your inference stack secure.", "source_url": "https://github.com/advisories/GHSA-h3rc-j82p-2c3h", "audio_url": "/content/cybersecurity/audio/2026-10-29/CVE-2026-115915.mp3", "layer": "ai", "exploited": "unknown", "exploited_source": "", "fix_available": true, "fix_version": "0.32.0", "ecosystem_match": 2, "blast_radius": 3, "severity": "high", "ghsa_id": "GHSA-h3rc-j82p-2c3h", "source": "GitHub Advisories (pip)", "published": "2026-10-28"}, {"priority_rank": 2, "cve_id": "CVE-2026-115912", "name": "vLLM: Public multimodal SHM handle forgery reaches worker-side `pickle.loads()` with insecure serialization disabled", "vendor": "", "product": "vllm", "cwe": ["CWE-502"], "epss": 0.00481, "priority_reason": "AI blast radius (pickle); no exploitation reported; patch available (0.32.0)", "narrative": "vLLM is also affected by a public multimodal shared memory handle forgery that can reach worker-side `pickle.loads()`. This occurs when insecure serialization is disabled, yet the system still processes untrusted data through the shared memory cache. The issue is rated high severity with a blast radius of 3 in the AI layer, and no exploitation has been reported so far. The EPSS is 0.00481, which is low but non-zero, indicating a theoretical risk that could be realized if an attacker can manipulate the shared memory handles. For platform engineers, this means any vLLM deployment handling multimodal data is potentially exposed to code execution via deserialization. The fix is available in version 0.32.0, so you should upgrade alongside the previous vLLM patch. This closes a subtle but dangerous path for attackers to inject malicious objects into your inference workers.", "source_url": "https://github.com/advisories/GHSA-6757-hggm-hp2w", "audio_url": "/content/cybersecurity/audio/2026-10-29/CVE-2026-115912.mp3", "layer": "ai", "exploited": "unknown", "exploited_source": "", "fix_available": true, "fix_version": "0.32.0", "ecosystem_match": 2, "blast_radius": 3, "severity": "high", "ghsa_id": "GHSA-6757-hggm-hp2w", "source": "GitHub Advisories (pip)", "published": "2026-10-28"}, {"priority_rank": 3, "cve_id": "CVE-2026-93327", "name": "GCP-2026-074: A Deserialization of Untrusted Data vulnerability was discovered in the SubWorkflow task in Application Integration versions prior to September 11, 2026.", "vendor": "", "product": "", "cwe": [], "epss": 0.0, "priority_reason": "cloud-account blast radius (gcp); no exploitation reported; patch available", "narrative": "Google Cloud has patched a deserialization of untrusted data vulnerability in the SubWorkflow task of Application Integration. This issue allowed an authenticated user to execute arbitrary code by submitting a maliciously crafted serialized object. It sits in the cloud_account layer with a blast radius of 4, which is wider than the AI-specific items but still contained to the service. No exploitation is reported, and the EPSS is 0.0, suggesting a low immediate risk. However, for teams using Application Integration for complex workflows, this is a significant trust boundary. The vulnerability was patched on September 11, 2026, so no customer action is required if you are on the latest version. This is a good reminder to keep your GCP services up to date, as deserialization flaws in cloud-native services can often lead to privilege escalation within the account.", "source_url": "https://docs.cloud.google.com/support/bulletins/index#gcp-2026-074", "audio_url": "/content/cybersecurity/audio/2026-10-29/CVE-2026-93327.mp3", "layer": "cloud_account", "exploited": "unknown", "exploited_source": "", "fix_available": true, "fix_version": "", "ecosystem_match": 2, "blast_radius": 4, "source": "Google Cloud security bulletins", "published": "2026-10-28"}, {"priority_rank": 4, "cve_id": "CVE-2026-118029", "name": "CVE-2026-118029 - Unverified ownership of the default Athena query result location in AWS SDK for pandas", "vendor": "", "product": "", "cwe": ["CWE-283"], "epss": 0.0, "priority_reason": "cloud-account blast radius (aws); no exploitation reported", "narrative": "The AWS SDK for pandas (awswrangler) has an unverified ownership issue in the default Athena query result location. This means that if you are using the library to run queries, the results might be written to a location that another user or process can access. It is a cloud_account layer issue with a blast radius of 4, and no exploitation is reported. The EPSS is 0.0, but the nature of the flaw—improper access control—means it could lead to data leakage in multi-tenant environments. For data engineers using awswrangler in their pipelines, this is a subtle risk that could expose sensitive query results. There is no specific fix version mentioned in the bulletin, so you should check for updates to the library or ensure that your S3 buckets have strict access controls. This is a 'keep an eye on it' item, as it requires careful configuration rather than a simple patch.", "source_url": "https://aws.amazon.com/security/security-bulletins/rss/2026-153-aws/", "audio_url": "/content/cybersecurity/audio/2026-10-29/CVE-2026-118029.mp3", "layer": "cloud_account", "exploited": "unknown", "exploited_source": "", "fix_version": "", "ecosystem_match": 2, "blast_radius": 4, "source": "AWS Security Bulletins", "published": "2026-10-28"}, {"priority_rank": 5, "cve_id": "CVE-2026-89049", "name": "CVE-2026-89049 - Server-side request forgery in the Session Manager port forwarding functionality in AWS Systems Manager Agent", "vendor": "", "product": "", "cwe": ["CWE-1289", "CWE-918"], "epss": 0.00662, "priority_reason": "cloud-account blast radius (aws); no exploitation reported", "narrative": "AWS Systems Manager Agent has a server-side request forgery (SSRF) vulnerability in its Session Manager port forwarding functionality. This allows an attacker to make requests to internal resources that should not be accessible from the agent. It is a cloud_account layer issue with a blast radius of 4, and no exploitation is reported. The EPSS is 0.00662, which is slightly higher than some other items, suggesting a moderate risk. For teams using SSM for remote access to EC2 instances, this could be a path to internal network reconnaissance. There is no fix version specified, so you should monitor for updates from AWS. In the meantime, ensure that your SSM endpoints are not exposed to untrusted networks and that you have strict security groups in place. This is a 'worth patching this week' item if you are running SSM agents in sensitive environments.", "source_url": "https://aws.amazon.com/security/security-bulletins/rss/2026-107-aws/", "audio_url": "/content/cybersecurity/audio/2026-10-29/CVE-2026-89049.mp3", "layer": "cloud_account", "exploited": "unknown", "exploited_source": "", "fix_version": "", "ecosystem_match": 2, "blast_radius": 4, "source": "AWS Security Bulletins", "published": "2026-10-25"}, {"priority_rank": 6, "cve_id": "CVE-2026-89090", "name": "CVE-2026-89090 - Denial of service in the event stream header decoder in AWS SDK for Go v2", "vendor": "", "product": "", "cwe": ["CWE-248"], "epss": 0.00305, "priority_reason": "cloud-account blast radius (aws); no exploitation reported", "narrative": "The AWS SDK for Go v2 has a denial of service vulnerability in its event stream header decoder. This could allow an attacker to crash the SDK by sending malformed headers, which would disrupt any application relying on it for streaming data. It is a cloud_account layer issue with a blast radius of 4, and no exploitation is reported. The EPSS is 0.00305, which is low, but the impact on availability could be significant for critical services. For developers using the Go SDK in their Kubernetes workloads, this is a risk that could lead to pod restarts or service outages. There is no fix version specified, so you should check for updates to the SDK. This is a 'keep an eye on it' item, as it is more of a stability issue than a security breach, but it is still worth addressing to ensure the resilience of your cloud-native applications.", "source_url": "https://aws.amazon.com/security/security-bulletins/rss/2026-110-aws/", "audio_url": "/content/cybersecurity/audio/2026-10-29/CVE-2026-89090.mp3", "layer": "cloud_account", "exploited": "unknown", "exploited_source": "", "fix_version": "", "ecosystem_match": 2, "blast_radius": 4, "source": "AWS Security Bulletins", "published": "2026-10-25"}, {"priority_rank": 7, "cve_id": "CVE-2026-115419", "name": "CVE-2026-115419 - Unintended disclosure of data element to wrong session in Strands Agents SDK A2A Server", "vendor": "", "product": "", "cwe": ["CWE-488"], "epss": 0.00426, "priority_reason": "cloud-account blast radius (aws); no exploitation reported", "narrative": "The Strands Agents SDK A2A Server has an issue where data elements can be disclosed to the wrong session. This means that if multiple agents are interacting with the server, one agent’s data could be leaked to another. It is a cloud_account layer issue with a blast radius of 4, and no exploitation is reported. The EPSS is 0.00426, which is low, but the impact on data confidentiality could be high. For teams building multi-agent systems with Strands, this is a risk that could lead to sensitive information being exposed. There is no fix version specified, so you should check for updates to the SDK. This is a 'keep an eye on it' item, as it requires careful session management and isolation to prevent data leakage. Ensure that your agents are running in isolated environments and that you have strict access controls in place.", "source_url": "https://aws.amazon.com/security/security-bulletins/rss/2026-146-aws/", "audio_url": "/content/cybersecurity/audio/2026-10-29/CVE-2026-115419.mp3", "layer": "cloud_account", "exploited": "unknown", "exploited_source": "", "fix_version": "", "ecosystem_match": 2, "blast_radius": 4, "source": "AWS Security Bulletins", "published": "2026-10-22"}, {"priority_rank": 8, "cve_id": "CVE-2026-105014", "name": "CVE-2026-105014 Azure Pipelines Remote Code Execution Vulnerability", "vendor": "", "product": "", "cwe": ["CWE-77"], "epss": 0.00667, "priority_reason": "cloud-account blast radius (azure); no exploitation reported", "narrative": "Azure Pipelines has a remote code execution vulnerability due to improper neutralization of special elements in a command. This allows an unauthorized attacker to execute code over a network. It is a cloud_account layer issue with a blast radius of 4, and no exploitation is reported. The EPSS is 0.00667, which is moderate, suggesting a realistic risk if the pipeline is exposed to untrusted input. For teams using Azure DevOps for CI/CD, this is a significant risk, as a compromised pipeline could lead to the deployment of malicious code. There is no fix version specified, so you should monitor for updates from Microsoft. In the meantime, ensure that your pipelines are not processing untrusted input without proper sanitization. This is a 'worth patching this week' item if you are using Azure Pipelines in a production environment.", "source_url": "https://msrc.microsoft.com/update-guide/vulnerability/CVE-2026-105014", "audio_url": "/content/cybersecurity/audio/2026-10-29/CVE-2026-105014.mp3", "layer": "cloud_account", "exploited": "unknown", "exploited_source": "", "fix_version": "", "ecosystem_match": 2, "blast_radius": 4, "source": "MSRC Security Update Guide", "published": "2026-10-22"}, {"priority_rank": 9, "cve_id": "CVE-2026-108456", "name": "CVE-2026-108456 Azure AI Foundry Elevation of Privilege Vulnerability", "vendor": "", "product": "", "cwe": ["CWE-918"], "epss": 0.00701, "priority_reason": "cloud-account blast radius (azure); no exploitation reported", "narrative": "Azure AI Foundry has a server-side request forgery (SSRF) vulnerability that allows an authorized attacker to elevate privileges over a network. This means that if an attacker has access to the AI Foundry service, they could potentially make requests to internal resources that should not be accessible. It is a cloud_account layer issue with a blast radius of 4, and no exploitation is reported. The EPSS is 0.00701, which is the highest in this list, suggesting a higher risk. For teams using Azure AI Foundry for model deployment and management, this is a significant risk, as a compromised service could lead to privilege escalation within the Azure account. There is no fix version specified, so you should monitor for updates from Microsoft. This is a 'patch today' item if you are using Azure AI Foundry in a sensitive environment.", "source_url": "https://msrc.microsoft.com/update-guide/vulnerability/CVE-2026-108456", "audio_url": "/content/cybersecurity/audio/2026-10-29/CVE-2026-108456.mp3", "layer": "cloud_account", "exploited": "unknown", "exploited_source": "", "fix_version": "", "ecosystem_match": 2, "blast_radius": 4, "source": "MSRC Security Update Guide", "published": "2026-10-22"}]}