AiSOC CVE-2026-103056

Exploitation requires reaching the AiSOC actions service and invoking a CrowdStrike RTR-backed endpoint action (quarantine_file/rm, the automatic path_exists/ls probe, or a runscript path) with attacker-controlled file_path, path, script_name, or script_args on an affected version (7.2.0 up to but excluding 12.0.0) with at least one CrowdStrike-managed endpoint enrolled. Per CVSS PR:L the baseline requirement is a low-privileged authenticated account; however, the source evidence shows that on default installs AISOC_ACTIONS_SERVICE_TOKEN was unset and the guard exempted itself whenever AISOC_DEV_MODE was enabled (the default), so mutating action routes responded to anyone able to reach the port - materially lowering the practical authentication barrier. The main limiting factors are the small number of AiSOC deployments and the need for the platform to actually manage CrowdStrike endpoints for the injected command to land.

This is a genuine high-priority flaw whose real-world weight is tempered mainly by product footprint, not by technical severity. The vendor CVSS 4.0 vector (AV:N/AC:L/AT:N/PR:L/UI:P) with full vulnerable- and subsequent-system impact (VC/VI/VA:H + SC/SI/SA:H, score 9.4) is well-justified: a low-privileged authenticated AiSOC user turns a self-service response action into arbitrary code execution on any CrowdStrike-managed endpoint at SYSTEM/root, so the blast radius is the entire endpoint fleet the SOAR platform can reach, not just AiSOC itself - this is the classic 'compromise the security tool, own everything it manages' pattern. No EPSS score, KEV listing, or SSVC decision data was provided; there is no structured POC=YES flag either, but the patch commit's own comments reference 'the three reports' and 'published payloads' that closed the single quote, meaning exploit detail is effectively public even if a packaged exploit is not confirmed. One important nuance from the source evidence: the same fix commit also closes a default authentication gap - AISOC_ACTIONS_SERVICE_TOKEN was never generated while the guard exempted itself whenever the token was empty and AISOC_DEV_MODE was set (the default), so the mutating actions routes 'answered 200 to anyone who could reach the port.' On such default/dev installs the effective privilege requirement is lower than PR:L implies, pushing real risk upward. The dominant limiter is deployment scale: AiSOC is an obscure open-source project, so this is a real priority for the organizations that run it rather than a broad ecosystem event.

A low-privileged analyst (or, on a default dev-mode install where the actions service token was never set, any unauthenticated party who can reach the port) submits a containment action - e.g. quarantine-file - with a file_path like foo'; <malicious command>; echo '. The unescaped value breaks out of the single-quoted RTR argument in crowdstrike_rtr.py, and the crafted command runs on the targeted managed endpoint at the CrowdStrike agent's SYSTEM/root privilege. Exploit payloads are referenced as already published in the patch commit, so the injection technique is public knowledge.