NousResearch — CraftedSignal Threat Feed

NousResearch hermes-agent Sandbox Vulnerability (CVE-2026-9368)

hello@craftedsignal.io — Tue, 26 May 2026 13:46:59 +0000

A remote code execution vulnerability, identified as CVE-2026-9368, exists in NousResearch hermes-agent versions up to 2026.4.16. The vulnerability resides within the execute_code function of the tools/code_execution_tool.py file, specifically affecting the Environment Variable Handler component. A publicly available exploit allows for remote attackers to bypass the intended sandbox restrictions. The vendor, NousResearch, was contacted but did not respond to the disclosure. This vulnerability poses a significant risk as it allows attackers to execute arbitrary code outside of the intended hermes-agent sandbox.

Attack Chain

Attacker identifies a vulnerable hermes-agent instance running a version up to 2026.4.16.
The attacker crafts a malicious request targeting the execute_code function in tools/code_execution_tool.py.
The request exploits the vulnerability in the Environment Variable Handler component.
The vulnerability allows the attacker to manipulate environment variables in a way that bypasses sandbox restrictions.
The attacker injects arbitrary code into the environment, leveraging the compromised environment variables.
The execute_code function executes the injected code, now running outside the intended sandbox.
The attacker gains unauthorized access to the underlying system.

Impact

Successful exploitation of CVE-2026-9368 allows a remote attacker to bypass the sandbox restrictions of hermes-agent, leading to arbitrary code execution on the host system. This can result in complete system compromise, data theft, or denial of service. The vulnerability is remotely exploitable and has a publicly available exploit, increasing the likelihood of exploitation.

Recommendation

Monitor network traffic for requests targeting the execute_code function in tools/code_execution_tool.py to detect potential exploitation attempts using the Sigma rule provided.
Implement input validation and sanitization for environment variables to mitigate the vulnerability in the Environment Variable Handler component.
Apply network segmentation to limit the impact of a successful sandbox escape.
Monitor process creation events for suspicious processes spawned by the hermes-agent process to detect potential post-exploitation activity using the Sigma rule provided.

NousResearch hermes-agent OS Command Injection Vulnerability (CVE-2026-9367)

hello@craftedsignal.io — Tue, 26 May 2026 13:46:43 +0000

A vulnerability, identified as CVE-2026-9367, exists in NousResearch hermes-agent up to version 5157f5427f19488b31c6fdebbacd15d798ce7f63. The vulnerability resides within the detect_dangerous_command function located in the tools/approval.py file of the terminal_tool component. This flaw enables a remote attacker to inject arbitrary operating system commands. Publicly available exploits exist, increasing the risk of exploitation. The vendor was notified about the vulnerability but has not responded. This vulnerability poses a significant risk to systems running vulnerable versions of hermes-agent, potentially allowing for complete system compromise.

Attack Chain

The attacker identifies a vulnerable instance of NousResearch hermes-agent running a version up to 5157f5427f19488b31c6fdebbacd15d798ce7f63.
The attacker crafts a malicious input designed to be processed by the detect_dangerous_command function.
The attacker sends this crafted input to the vulnerable terminal_tool component.
The detect_dangerous_command function fails to properly sanitize the input, allowing the injection of OS commands.
The injected OS command is executed by the system with the privileges of the hermes-agent process.
The attacker gains arbitrary code execution on the target system.
The attacker may then install malware, exfiltrate sensitive data, or pivot to other systems within the network.

Impact

Successful exploitation of CVE-2026-9367 allows a remote, unauthenticated attacker to execute arbitrary commands on the affected system. This can lead to a complete compromise of the system, including the theft of sensitive information, installation of malware, and potential lateral movement within the network. Given the nature of the hermes-agent as an agent, this vulnerability could potentially expose numerous systems if successfully exploited.

Recommendation

Apply any available patches or updates provided by NousResearch to address CVE-2026-9367.
Monitor network traffic for suspicious commands being sent to systems running hermes-agent. Deploy the provided Sigma rule Detect Hermes-Agent Command Injection via detect_dangerous_command to identify command injection attempts.
Implement input validation and sanitization measures within the detect_dangerous_command function to prevent OS command injection.
Review and restrict the permissions of the hermes-agent process to minimize the impact of successful exploitation.

NousResearch hermes-agent Injection Vulnerability (CVE-2026-9366)

hello@craftedsignal.io — Tue, 26 May 2026 13:46:26 +0000

A vulnerability, identified as CVE-2026-9366, has been discovered in NousResearch hermes-agent version 2026.4.23. This injection vulnerability resides within the _scan_context_content function located in the agent/prompt_builder.py file. The vulnerability can be exploited remotely, and publicly available exploits exist. The vendor was contacted about the disclosure but did not respond. This vulnerability matters because it allows attackers to inject malicious code into the application potentially leading to arbitrary code execution.

Attack Chain

Attacker identifies a vulnerable instance of NousResearch hermes-agent running version 2026.4.23.
Attacker crafts a malicious input string designed to exploit the injection vulnerability in the _scan_context_content function.
Attacker sends the crafted input to the vulnerable function, potentially through a network request or API call.
The _scan_context_content function fails to properly neutralize special elements within the input, leading to code injection.
The injected code is executed within the context of the hermes-agent application.
The attacker gains control over parts of the application.
The attacker escalates privileges within the application.
The attacker achieves arbitrary code execution on the server.

Impact

Successful exploitation of CVE-2026-9366 can allow an attacker to execute arbitrary code on the server running NousResearch hermes-agent. The affected version is 2026.4.23. Given the nature of injection vulnerabilities, it’s plausible that attackers could leverage this to gain full control over the system, potentially leading to data breaches, service disruption, or further lateral movement within the network.

Recommendation

Upgrade NousResearch hermes-agent to a patched version that addresses CVE-2026-9366 (no version available).
Implement input validation and sanitization for the _scan_context_content function in agent/prompt_builder.py to prevent injection attacks.
Monitor network traffic for suspicious patterns and payloads targeting the hermes-agent application.
Deploy the Sigma rules provided below to your SIEM to detect potential exploitation attempts of CVE-2026-9366.
Enable and review application logs for anomalies related to the _scan_context_content function.

NousResearch hermes-agent Remote Code Injection Vulnerability (CVE-2026-9353)

hello@craftedsignal.io — Tue, 26 May 2026 13:44:43 +0000

A security vulnerability, CVE-2026-9353, has been identified in NousResearch hermes-agent, affecting versions up to 2026.4.23. The vulnerability resides in the agent/skills_guard.py file within the Skills Guard Multi-Word Prompt Handler component. By manipulating the THREAT_PATTERNS argument, a remote attacker can inject arbitrary code. Public disclosure of the exploit is available, increasing the risk of exploitation. The vendor was contacted regarding the vulnerability, but no response was received. This vulnerability allows for unauthenticated remote code execution, potentially leading to complete system compromise.

Attack Chain

An attacker identifies a vulnerable NousResearch hermes-agent instance running a version prior to 2026.4.23.
The attacker crafts a malicious payload designed to exploit the code injection vulnerability in the THREAT_PATTERNS argument.
The attacker sends a specially crafted request to the hermes-agent server, embedding the malicious payload within the THREAT_PATTERNS argument targeting the Skills Guard Multi-Word Prompt Handler.
The hermes-agent server processes the request, failing to properly sanitize or validate the THREAT_PATTERNS input.
Due to insufficient input validation, the malicious payload is injected and executed by the server.
The injected code allows the attacker to execute arbitrary commands on the server, potentially gaining shell access.
The attacker leverages the compromised server to perform further actions, such as data exfiltration or lateral movement within the network.
The attacker achieves complete system compromise and gains persistent access to the target environment.

Impact

Successful exploitation of CVE-2026-9353 can lead to remote code execution on the affected NousResearch hermes-agent server. This could allow an attacker to gain complete control over the system, potentially leading to data breaches, service disruption, or further attacks on the internal network. Given the public availability of the exploit, the likelihood of exploitation is increased, posing a significant risk to organizations using vulnerable versions of hermes-agent.

Recommendation

Upgrade NousResearch hermes-agent to a version later than 2026.4.23 to remediate CVE-2026-9353.
Deploy the Sigma rule “Detect CVE-2026-9353 Exploitation Attempt via Malicious THREAT_PATTERNS Argument” to detect potential exploitation attempts by monitoring HTTP requests for suspicious patterns.
Implement input validation and sanitization measures on all user-supplied inputs to prevent code injection vulnerabilities.
Monitor network traffic for unusual activity originating from the hermes-agent server.

NousResearch hermes-agent Missing Authorization Vulnerability (CVE-2026-9350)

hello@craftedsignal.io — Tue, 26 May 2026 13:44:22 +0000

A missing authorization vulnerability, identified as CVE-2026-9350, affects NousResearch hermes-agent up to version 2026.4.16. The flaw resides within the check_all_command_guards function in the tools/approval.py file, a part of the Batch Runner component. Successful exploitation of this vulnerability allows remote attackers to bypass authorization checks, potentially leading to unauthorized command execution or data access. Publicly available exploit code exists, increasing the risk of exploitation. The vendor was notified but has not responded. This vulnerability was published on 2026-05-24.

Attack Chain

Attacker identifies a vulnerable hermes-agent instance running a version prior to 2026.4.16.
The attacker crafts a malicious request targeting the Batch Runner component.
The request is designed to invoke functionality that relies on the check_all_command_guards function.
Due to the missing authorization check, the attacker bypasses the intended access controls.
The attacker executes unauthorized commands or accesses restricted data within the Batch Runner component.
The attacker leverages the unauthorized access to escalate privileges or gain further control over the system.
The attacker may install malware, exfiltrate sensitive data, or disrupt services.

Impact

Successful exploitation of CVE-2026-9350 can lead to unauthorized access to sensitive data, execution of arbitrary commands, and potential system compromise. This can result in data breaches, service disruption, and reputational damage. The fact that the exploit is publicly available increases the likelihood of widespread exploitation.

Recommendation

Upgrade NousResearch hermes-agent to a version later than 2026.4.16 to remediate CVE-2026-9350.
Monitor network traffic for suspicious requests targeting the Batch Runner component of hermes-agent using the “Detect Suspicious Hermes-Agent Batch Runner Requests” Sigma rule.
Deploy the “Detect Unauthorized Command Execution via Hermes-Agent” Sigma rule to identify potential exploitation attempts.