<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Model-Loading — PoC Archive</title><link>https://poc.intelseclab.com/tags/model-loading/</link><description>Latest proof-of-concept entries.</description><generator>Hugo</generator><language>en-us</language><lastBuildDate>Sun, 05 Jul 2026 00:00:00 +0000</lastBuildDate><atom:link href="https://poc.intelseclab.com/tags/model-loading/index.xml" rel="self" type="application/rss+xml"/><item><title>NVIDIA Triton Inference Server SageMaker Auth Bypass to Unauthenticated RCE (CVE-2026-24207)</title><link>https://poc.intelseclab.com/pocs/network/2026-07-05_cve-2026-24207-triton-sagemaker-auth-bypass-rce/</link><pubDate>Sun, 05 Jul 2026 00:00:00 +0000</pubDate><guid>https://poc.intelseclab.com/pocs/network/2026-07-05_cve-2026-24207-triton-sagemaker-auth-bypass-rce/</guid><description>Critical severity (CVSS 9.8) — network · CVE-2026-24207 (sibling: CVE-2026-24206, Vertex AI, analysis only). Status: Weaponized. Affects: NVIDIA Triton Inference Server. Tags: nvidia-triton, sagemaker, auth-bypass, rce, cwe-288, ml-inference, model-loading, pre-auth.</description><category>network</category><category>Critical</category><category>nvidia-triton</category><category>sagemaker</category><category>auth-bypass</category><category>rce</category><category>cwe-288</category><category>ml-inference</category><category>model-loading</category><category>pre-auth</category></item><item><title>AI Model-Loader `trust_remote_code` Order-of-Operations RCE Simulation (CVE-2026-22807)</title><link>https://poc.intelseclab.com/pocs/misc/2026-07-05_cve-2026-22807-vllm-trust-remote-code-bypass/</link><pubDate>Sun, 05 Jul 2026 00:00:00 +0000</pubDate><guid>https://poc.intelseclab.com/pocs/misc/2026-07-05_cve-2026-22807-vllm-trust-remote-code-bypass/</guid><description>High severity — misc · CVE-2026-22807. Status: PoC. Affects: AI inference/model-loading frameworks that resolve custom model classes via auto_map before validating trust_remote_code (pattern seen in vLLM/Transformers-style loaders). Tags: vllm, huggingface, trust-remote-code, toctou, model-loading, supply-chain, python, code-execution.</description><category>misc</category><category>High</category><category>vllm</category><category>huggingface</category><category>trust-remote-code</category><category>toctou</category><category>model-loading</category><category>supply-chain</category><category>python</category><category>code-execution</category></item></channel></rss>