Loading vulnerability…
CVE-2026-2256 is a command injection vulnerability affecting ModelScope ms-agent versions 1.6.0rc1 and earlier. According to the provided content, the flaw allows attacker-controlled, prompt-derived input to reach an operating system command execution path without sufficient neutralization of special elements. As a result, a crafted prompt or prompt-influenced input can cause the application to execute arbitrary OS commands on the host running ms-agent.
Mallory correlates every CVE against your assets, your vendors, and active adversary campaigns. Know which vulnerabilities matter for you, not just which ones are loud.
What it means. What to do now. Patch path, mitigations, and the assume-compromise checklist.
What an attacker gets, and what they’ve been doing with it.
If you can’t patch tonight, do this now.
Patch, then assume compromise.
1 valid exploit after Mallory filtered fakes, detection scripts, and README-only repos (2 hidden).
Repository contains a single Python PoC (reverse_shell_poc.py), a README, and an Apache-2.0 LICENSE. The README claims CVE-2026-2256: a command-injection / unsafe command execution issue in ModelScope MS-Agent v1.5.2, specifically the Shell tool’s check_safe() validation (ms_agent/tools/shell/shell.py). The PoC re-implements check_safe() as a blocklist-based validator and demonstrates that a reverse-shell payload can pass validation because interpreters like python3 are not blocked. Core capability: bypass of a naive command blocklist to achieve arbitrary command execution; when run with --execute, it launches a Python one-liner that opens a TCP connection to 127.0.0.1:1111, dup2’s the socket to stdin/stdout/stderr, and spawns /bin/sh -i, yielding an interactive reverse shell to a local netcat listener. Structure/purpose: - README.md: explains the alleged vulnerability, affected product/version, and step-by-step local testing using nc -l 1111 and ./reverse_shell_poc.py --execute. - reverse_shell_poc.py: (1) defines check_safe() with regex-based dangerous command/path/redirection/env-var checks; (2) constructs a python3 reverse shell command that avoids the blocklist; (3) validates it and optionally executes it via subprocess.run(..., shell=True, cwd=/tmp/test_output). Notable targeting/constraints: the PoC is explicitly localhost-only (connects to 127.0.0.1) to reduce abuse risk, but the technique generalizes to any environment where the Shell tool executes user-influenced commands and relies on incomplete blocklisting.
Products and vendors Mallory has correlated with this vulnerability. Open in Mallory to drill down to specific CPE configurations and version ranges.
Vendor-confirmed product mapping. Mallory continuously reconciles this list against your asset inventory.
18 sources tracked across advisories and community write-ups. News coverage will land here when it surfaces.
No news coverage yet. Advisories and community discussion only.
Query your assets running an affected version, and investigate the blast radius.
Every observed campaign linking this CVE to a named adversary.
Malware families riding this exploit, with evidence and IOCs.
YARA, Sigma, Snort, and vendor rules, auto-deployed to your SIEM.
Cross-references every affected SKU, including bundled OEM variants.
Community discussion across Reddit, Mastodon, and other social sources.