Network-AI versions 5.14.0‑5.15.1 contain a vulnerability in the ClaudeHookBridge PreToolUse hook that applies deny patterns to only the first candidate field. An attacker controlling a prompt‑injected agent can insert malicious payloads into fields such as Write content or an MCP script parameter alongside harmless values like file_path or command, thereby evading the deny patterns and writing web shells or executing arbitrary scripts. This flaw permits attackers to gain execution privileges within the affected system, compromising confidentiality, integrity, and availability of applications running on Network-AI.

The affected product is Jovancoding’s Network-AI software. Versions from 5.14.0 up to, but not including, 5.15.2 are vulnerable. Systems running any of these releases should be reviewed for potential exploitation.

The CVSS score of 7.5 indicates a high severity of this denial‑pattern bypass. Although no EPSS score is available, the exploitability is considered high because the attacker only needs to influence the agent’s prompt content, a scenario common in many AI‑assistant deployments. The vulnerability is not listed in the CISA KEV catalog, but its high CVSS and potential for arbitrary code execution make it a priority for rapid remediation. The likely attack vector is local to systems that host the Network‑AI agent, requiring the attacker to supply crafted prompt inputs to the agent process.