Loading vulnerability details...
| Property | Value |
|---|---|
| Advisory ID | SGZ-2026-91FAF |
| Severity | HIGH |
| CWE | CWE-78 |
| Confidence | 78% |
| Category | injection |
| Exploitability | possible |
| Package | ACE_TAO@latest (github) |
| Location | ACE/bin/make_release.py:78-82 |
| Attack Vector | NETWORK |
| Attack Complexity | LOW |
| Privileges Required | LOW |
| Discovered By | SafeGuard Zero-Day AI Discovery Engine |
The script defines an ex helper that executes arbitrary shell commands using subprocess without proper sanitization or argument separation. If an attacker can influence the command argument (e.g., via manipulated CLI options that are later passed to ex), they could execute arbitrary commands on the host running the release script.
The vulnerability was identified in the file ACE/bin/make_release.py at lines 78-82 within the ACE_TAO package (version latest). Any code path that invokes these functions inherits this vulnerability.
File: ACE/bin/make_release.py (lines 78-82)
def ex (command, allow_fail=False):
if vprint ("Executing", command, take_action=True):
return
# ... likely uses subprocess.run(command, shell=True) ...
The code above demonstrates the vulnerable pattern. This code is executed at runtime and can be directly exploited by an attacker with the appropriate access level.
Supply a crafted value to a CLI argument that later gets concatenated into the command string passed to ex, causing the subprocess to execute attacker-controlled commands.
Remote code execution on the build host, potentially compromising the CI environment and any secrets stored there.
Replace the use of subprocess with argument list form (subprocess.run([...], check=True)) and avoid shell=True. Validate and whitelist any inputs that become part of the command string.
Advisory: SGZ-2026-91FAF | Source: SafeGuard Zero-Day AI Discovery | Status: Candidate
This vulnerability was autonomously discovered by SafeGuard's AI-powered Zero-Day Discovery engine using TAOR (Think-Act-Observe-Repeat) agentic analysis on the package source code. It is not yet tracked in any public vulnerability database (CVE, NVD, GHSA, OSV). This finding should be triaged by a security engineer and, if confirmed, reported upstream to the package maintainer.
| Property | Value |
|---|---|
| Advisory ID | SGZ-2026-91FAF |
| Severity | HIGH |
| CWE | CWE-78 |
| Confidence | 78% |
| Category | injection |
| Exploitability | possible |
| Package | ACE_TAO@latest (github) |
| Location | ACE/bin/make_release.py:78-82 |
| Attack Vector | NETWORK |
| Attack Complexity | LOW |
| Privileges Required | LOW |
| Discovered By | SafeGuard Zero-Day AI Discovery Engine |
The script defines an ex helper that executes arbitrary shell commands using subprocess without proper sanitization or argument separation. If an attacker can influence the command argument (e.g., via manipulated CLI options that are later passed to ex), they could execute arbitrary commands on the host running the release script.
The vulnerability was identified in the file ACE/bin/make_release.py at lines 78-82 within the ACE_TAO package (version latest). Any code path that invokes these functions inherits this vulnerability.
File: ACE/bin/make_release.py (lines 78-82)
def ex (command, allow_fail=False):
if vprint ("Executing", command, take_action=True):
return
# ... likely uses subprocess.run(command, shell=True) ...
The code above demonstrates the vulnerable pattern. This code is executed at runtime and can be directly exploited by an attacker with the appropriate access level.
Supply a crafted value to a CLI argument that later gets concatenated into the command string passed to ex, causing the subprocess to execute attacker-controlled commands.
Remote code execution on the build host, potentially compromising the CI environment and any secrets stored there.
Replace the use of subprocess with argument list form (subprocess.run([...], check=True)) and avoid shell=True. Validate and whitelist any inputs that become part of the command string.
Advisory: SGZ-2026-91FAF | Source: SafeGuard Zero-Day AI Discovery | Status: Candidate
This vulnerability was autonomously discovered by SafeGuard's AI-powered Zero-Day Discovery engine using TAOR (Think-Act-Observe-Repeat) agentic analysis on the package source code. It is not yet tracked in any public vulnerability database (CVE, NVD, GHSA, OSV). This finding should be triaged by a security engineer and, if confirmed, reported upstream to the package maintainer. This vulnerability involves weaknesses in
This high-severity vulnerability could allow attackers to gain unauthorized access, execute arbitrary code, or compromise data integrity. Prompt remediation is strongly recommended.
Isolate affected systems from untrusted networks until patching is complete
Implement enhanced monitoring for exploitation attempts and unusual behavior