Loading vulnerability…
CVE-2026-25130 is an OS command injection vulnerability affecting Cybersecurity AI (CAI) framework versions through 0.5.10. Multiple function tools pass user-controlled values to subprocess execution with shell invocation enabled. The find_file() tool is particularly vulnerable because the underlying find command is treated as pre-approved and does not require human approval. An attacker can supply malicious find arguments, including execution-capable options, through the tool's args parameter and cause arbitrary host commands to run, bypassing CAI human-in-the-loop controls.
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.
Repository contains a Python proof-of-concept for CVE-2026-25130 affecting cai-framework (Cybersecurity AI / CAI Framework) <= 0.5.10. Structure is minimal: (1) a single PoC script "Cybersecurity AI (CAI) Framework.py" and (2) a README describing the bug, impact, and mitigation. Core capability: the script imports cai.tools.reconnaissance.filesystem.find_file and passes a malicious, user-controlled 'args' string intended for a filesystem search. The PoC assumes find_file builds an underlying system 'find' command unsafely (string concatenation / insufficient validation), enabling argument injection. By injecting the find flag '-exec', the attacker can execute arbitrary OS commands in the context of the running process. Demonstrated payloads: - Test 1 executes 'whoami' via "-type f -exec whoami \;" and prints returned output. - Test 2 creates a non-destructive marker file /tmp/CAI_CVE_2026_25130_PROOF via "-exec touch ...". - Test 3 (commented out) shows a reverse shell using bash TCP redirection to 127.0.0.1:4444. Attack vector: primarily local execution, but the README highlights higher risk in AI-agent/automation workflows where tool calls may be triggered without explicit human approval. Patch is referenced as occurring after commit e22a1220. Note: README’s example filename (cai_find_file_rce_poc.py) does not match the actual script name in the repository.
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.
8 sources tracked across advisories, community write-ups, and news. New activity surfaces here as Mallory finds it.
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.