CVE WatchOct 13, 2026

CVE-2026-107717

Banks generates meaningful LLM prompts using a simple template language. Prior to 2.5.0, Banks Prompt.chat_messages() attempts to parse ever

CVSS

6.5

Medium

EPSS

0.3%

p27

KEV

—

Exploit Today

0-100

Published: Oct 8, 2026 · Last modified: Oct 13, 2026 · CWE-20

EPSS · 30d

0.3%EPSS · 30 days0.3%

2026-10-092026-11-05

Technical description

Banks generates meaningful LLM prompts using a simple template language. Prior to 2.5.0, Banks Prompt.chat_messages() attempts to parse every line of rendered template output as ChatMessage JSON. When an application renders untrusted data and passes the returned ChatMessage objects to an LLM provider, attacker-controlled JSON can cross the prompt boundary and become a system, assistant, or tool message because ChatMessage.role accepts arbitrary strings. This can override application instructions, alter the intended prompt structure, or confuse downstream tool and message handling. This issue is fixed in version 2.5.0.

Official references

Related CVEs

CVECVSSEPSSKEVRExploitTitleMod.

CVE-2026-978486.5 MED

—

———A flaw was found in the token sanitization logic of Keycloak, an open source identity and access management solution. The issue occurs when a protocol mapper is configured to emit structured claims using the JSON type. The security check designed to remove unauthorized role mappings only looks for standard Java maps and fails to recognize Jackson based JSON objects. This allows an attacker to inject forged client roles into security tokens, which could lead to unauthorized access to protected resources.4h CVE-2026-487059.3 CRI

—

———MONAI (Medical Open Network for AI) is an AI toolkit for health care imaging. Prior to 1.6.1, PydicomReader._get_affine() trusts non-finite values in the DICOM PixelSpacing, ImagePositionPatient, and ImageOrientationPatient fields and composes them into an affine matrix without validation. An unauthenticated attacker who can upload a crafted DICOM to a default MONAILabel deployment can propagate NaN or infinity values through LoadImaged, Spacingd, model inference, and output writing, causing inference failures, persistent datastore poisoning, or geometrically corrupted results. This issue is fixed in version 1.6.1.17h CVE-2026-463509.1 CRI

—

———dcmjs is a JavaScript implementation of DICOM manipulation. Prior to 0.52.0, the DecimalString and IntegerString implementations in src/ValueRepresentation.js sanitize text with a permissive regular expression and pass the result to JavaScript's Number constructor without validating finiteness or the DICOM value-representation rules. Scientific notation overflow in a DS or IS field can produce positive or negative Infinity, while stripping alphabetic characters from signed infinity text can leave a sign-only string that converts to NaN. IntegerString also accepts exponent notation and values outside the DICOM signed 32-bit IS range. Applications that trust these parsed values can propagate invalid numbers into spatial geometry, display windowing, dose scaling, frame counts, rescaling, or volume reconstruction, causing incorrect calculations or disrupted processing. This issue is fixed in version 0.52.0.18h CVE-2026-715488.8 HIG

—

——0SuiteCRM is an open-source, enterprise-ready Customer Relationship Management (CRM) software application. Prior to 7.15.2 and 8.10.2, AOR_Report::build_report_query_where incorporates a report editor's parameter_value when value_type=Field without sufficient validation. An authenticated user who can save a report can use SQL injection to read the entire SuiteCRM database. Disclosed data can include users.user_hash, OAuth2 tokens, CRM records, report definitions, and configuration data; the connector does not permit stacked queries, so direct write-through-injection is not supported. This issue is fixed in versions 7.15.2 and 8.10.2.20h CVE-2026-1164722.7 LOW

—

——0A security issue was discovered in ingress-nginx where a user with permission to create an Ingress may be able to cause denial of service of the ingress-nginx data plane, or possibly redirect traffic from other Ingresses. Kubernetes clusters are only affected if they are using ingress-nginx. Due to a missing check, the mirror-target Ingress annotation can be used to target an unauthenticated configuration API within the ingress-nginx controller, allowing write commands to be submitted. Corrupting the data within that configuration API results in all traffic through ingress-nginx being dropped until an Ingress configuration change or a controller restart causes the data to be recreated correctly. In principle, this write access could also be used to redirect traffic from other Ingresses controlled by the same ingress-nginx instance. In practice, there is negligible additional risk: such redirection would require privileged information about internal ingress-nginx state, and users can already redirect traffic from other Ingresses controlled by the same ingress-nginx anyway.17h CVE-2026-971746.8 MED

—

——0A flaw was found in the Client-Initiated Backchannel Authentication (CIBA) implementation of Keycloak. The CIBA endpoint fails to properly validate the sub parameter, allowing it to be stored as an additional request parameter. During the token issuance process, this injected value overwrites the legitimate subject identifier. This allows an attacker to authenticate as themselves but receive security tokens belonging to a different victim user, leading to unauthorized account access.4h