Remote Command Execution in GPT Researcher via malicious MCP STDIO configuration
CVE-2025-65720 is a remote command execution vulnerability affecting GPT Researcher. Based on the provided reporting, the issue is tied to unsafe handling of Model Context Protocol (MCP) STDIO server configuration, where attacker-controlled command and argument values can be executed on the host running GPT Researcher. The supplied context states that all versions are affected and that exploitation can be triggered when a victim accesses an attacker-controlled crafted HTML page, which then causes command execution and can spawn a reverse shell on the machine running GPT Researcher locally. The vulnerability is described as part of a broader class of MCP STDIO command-injection issues in which untrusted configuration or content influences subprocess execution without sufficient sanitization or trust separation.
Are you exposed to this one?
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.
Impact, mitigation & remediation
What it means. What to do now. Patch path, mitigations, and the assume-compromise checklist.
Impact
What an attacker gets, and what they’ve been doing with it.
Mitigation
If you can’t patch tonight, do this now.
Remediation
Patch, then assume compromise.
Exploits
No public exploits tracked yet. Mallory keeps watching.
No public exploit code observed for this vulnerability.
Recent activity
5 sources tracked across advisories, community write-ups, and news. New activity surfaces here as Mallory finds it.
A CVE assigned to GPT Researcher in the context of stdio MCP or code-execution behavior in hosted single-user AI web services.
A vulnerability in GPT Researcher stemming from unsafe MCP STDIO configuration defaults that can enable command execution on the server.
A command injection vulnerability affecting GPT Researcher due to unsafe MCP STDIO behavior, potentially enabling arbitrary command execution and total system compromise.
A command injection vulnerability affecting GPT Researcher due to unsafe MCP STDIO behavior, potentially enabling arbitrary OS command execution and total system compromise.
The version that knows your environment.
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.