{"description":"Trending threats, MITRE ATT\u0026CK coverage, and detection metadata. Fed continuously.","favicon":"https://feed.craftedsignal.io/favicon-32x32.png","feed_url":"https://feed.craftedsignal.io/vendors/hank-ai/feed.json","home_page_url":"https://feed.craftedsignal.io/","icon":"https://feed.craftedsignal.io/apple-touch-icon.png","items":[{"_cs_actors":[],"_cs_cpes":[],"_cs_cves":[{"cvss":7.8,"id":"CVE-2026-72852"}],"_cs_exploited":false,"_cs_has_poc":false,"_cs_poc_references":[],"_cs_products":["darknet"],"_cs_severities":["high"],"_cs_tags":[],"_cs_type":"advisory","_cs_vendors":["hank-ai"],"content_html":"\u003cp\u003eThe Darknet framework is vulnerable to an integer overflow during the initialization of convolutional layers, tracked as CVE-2026-72852. The vulnerability exists within \u003ccode\u003esrc-lib/convolutional_layer.cpp\u003c/code\u003e, where heap buffer sizing for weights and outputs is calculated by multiplying configuration fields from a provided \u003ccode\u003e.cfg\u003c/code\u003e file using 32-bit integer arithmetic. If the product of these dimensions exceeds the \u003ccode\u003eINT_MAX\u003c/code\u003e limit, the integer wraps around, leading to the allocation of an undersized buffer.\u003c/p\u003e\n\u003cp\u003eAn attacker can trigger this vulnerability by providing a crafted \u003ccode\u003e.cfg\u003c/code\u003e file that specifies dimensions (such as width, height, and number of filters) that force an integer wrap-around. When \u003ccode\u003eforward_convolutional_layer\u003c/code\u003e processes the layer, it re-calculates dimensions and performs memory operations (reads and writes) based on the full expected size, while the underlying buffer remains undersized. This results in out-of-bounds memory access, including heap buffer overflow reads and potential heap metadata corruption, which can lead to application crashes or arbitrary code execution.\u003c/p\u003e\n\u003ch2 id=\"impact\"\u003eImpact\u003c/h2\u003e\n\u003cp\u003eSuccessful exploitation allows for heap memory corruption by simply loading a malicious configuration file for inference or training. This impact is significant for environments where untrusted models or configurations are processed, as it can lead to remote code execution or complete system compromise depending on the context in which the Darknet application is running.\u003c/p\u003e\n\u003ch2 id=\"recommendation\"\u003eRecommendation\u003c/h2\u003e\n\u003cp\u003ePrioritize patching or updating the Darknet framework to a version that implements safe integer arithmetic (such as checking for overflows before memory allocation) in \u003ccode\u003esrc-lib/convolutional_layer.cpp\u003c/code\u003e. Teams managing Darknet-based workloads should restrict the ability to load configuration files from untrusted sources.\u003c/p\u003e\n","date_modified":"2026-08-20T19:19:27Z","date_published":"2026-08-20T19:19:27Z","id":"https://feed.craftedsignal.io/briefs/2026-08-darknet-heap-overflow/","summary":"An integer overflow vulnerability in the Darknet neural network framework's convolutional layer initialization allows for heap buffer under-allocation and subsequent out-of-bounds memory access via malicious configuration files.","title":"Integer Overflow in Darknet Convolutional Layer Initialization","url":"https://feed.craftedsignal.io/briefs/2026-08-darknet-heap-overflow/"}],"language":"en","title":"CraftedSignal Threat Feed - Hank-Ai","version":"https://jsonfeed.org/version/1.1"}