Python Core

eval() executes arbitrary Python expressions from strings, enabling remote code execution when called with untrusted input.

exec() executes arbitrary Python statements from strings or code objects, enabling remote code execution when called with untrusted input.

code.InteractiveConsole and code.interact() enable arbitrary Python code execution and should not be exposed to untrusted users.

globals() exposes the module's global namespace as a mutable dictionary, allowing arbitrary attribute injection when passed to untrusted code.

__import__() or importlib.import_module() with a non-literal argument can import arbitrary modules when called with untrusted input.

typing.get_type_hints() evaluates string annotations as Python expressions, enabling code execution when annotation strings contain untrusted content.

os.system() and os.popen() execute shell commands via /bin/sh, enabling command injection when arguments contain untrusted input.

os.exec*() replaces the current process image with a new program, enabling arbitrary program execution when arguments are untrusted.

os.spawn*() spawns a new process and can execute arbitrary programs when the executable path or arguments are derived from untrusted input.

os.system() calls containing wildcard characters (*) may lead to unintended file inclusion or command injection through wildcard expansion.

Reverse shell pattern detected using socket connections with subprocess or exec(). This is a strong indicator of malicious code or a backdoor.

subprocess calls detected. Ensure command arguments are not user-controlled to prevent OS command injection.

subprocess called with shell=True passes the command through the system shell, enabling command injection when any part of the command contains untrusted input.

asyncio.create_subprocess_shell() passes the command through the system shell, enabling command injection when arguments contain untrusted input.

subinterpreters.run_string() executes arbitrary Python code strings in sub-interpreters, enabling code injection when called with untrusted input.

MD5 is cryptographically broken and unsuitable for security-sensitive purposes. Use SHA-256 or SHA-3 instead.

SHA-1 is cryptographically weak due to practical collision attacks. Use SHA-256 or SHA-3 for security-sensitive hashing.

hashlib.new() with an insecure algorithm name (MD5, SHA1, SHA-224) creates a cryptographically weak hash. Use SHA-256 or SHA-3.

SHA-224 and SHA3-224 provide only 112-bit collision resistance, which is below the 128-bit minimum recommended by NIST for new applications.

MD5 is being used to hash passwords. MD5 is cryptographically broken and orders of magnitude too fast for password hashing. Use bcrypt, scrypt, or argon2.

pickle.loads() and pickle.load() execute arbitrary Python code during deserialization. Never unpickle data from untrusted sources.

yaml.load() and yaml.unsafe_load() can execute arbitrary Python objects during YAML parsing. Use yaml.safe_load() instead.

jsonpickle.decode() can execute arbitrary Python code during deserialization. Use the standard json module for untrusted data.

ruamel.yaml configured with typ='unsafe' can instantiate arbitrary Python objects during YAML parsing. Use typ='safe' or the default round-trip loader.

marshal.loads() and marshal.load() are not secure against erroneous or malicious data and should not be used to deserialize untrusted input.

shelve.open() uses pickle internally for value serialization and is not safe for storing or retrieving data from untrusted sources.

dill.loads() and dill.load() extend pickle with broader serialization capabilities and can execute arbitrary code when deserializing untrusted data.

ssl._create_unverified_context() disables certificate verification entirely, making TLS connections vulnerable to man-in-the-middle attacks.

SSLContext configured with SSLv2, SSLv3, TLSv1.0, or TLSv1.1 uses deprecated protocols with known vulnerabilities. Use TLS 1.2 or TLS 1.3.

ssl.wrap_socket() is deprecated since Python 3.7 and should be replaced with SSLContext.wrap_socket() for proper TLS configuration.

TLS certificate validation is explicitly disabled via verify=False or CERT_NONE, making connections vulnerable to man-in-the-middle attacks.

http.client.HTTPConnection transmits data in plaintext without encryption. Use HTTPSConnection for sensitive communications.

HTTP URLs in requests calls transmit data in plaintext without encryption. Use HTTPS URLs for sensitive data transmission.

urllib.request.urlopen() over HTTP transmits data in plaintext. Verify HTTPS URLs are used and SSL context is properly configured.

urllib.request.Request() and OpenerDirector used with HTTP URLs transmit data in plaintext. Verify HTTPS URLs are used.

ftplib.FTP() transmits data and credentials in plaintext. Use ftplib.FTP_TLS() or SFTP (via paramiko) for secure file transfer.

telnetlib.Telnet() transmits all data including credentials in plaintext. Replace with SSH (paramiko) for remote command execution.

Binding a socket to 0.0.0.0 exposes the service on all network interfaces, including public-facing ones. Bind to specific interfaces in production.

paramiko.AutoAddPolicy() and WarningPolicy() automatically accept unknown SSH host keys, enabling man-in-the-middle attacks on SSH connections.

paramiko exec_command() runs commands on a remote host. Audit that command arguments are not derived from untrusted input to prevent command injection.

multiprocessing.Connection.recv() uses pickle internally and is not safe for receiving data from untrusted connections.

SQL query built with string formatting passed to psycopg2 cursor.execute() enables SQL injection. Use parameterized queries with %s placeholders.

SQL query built with string formatting detected. String-formatted SQL strings are a common SQL injection vector. Use parameterized queries.

xml.etree.ElementTree is vulnerable to XML External Entity (XXE) attacks. Use defusedxml for safe XML parsing.

xml.dom.minidom is vulnerable to XML External Entity (XXE) attacks. Use defusedxml.minidom for safe XML parsing.

xmlrpc.client.ServerProxy and xmlrpc.server modules are vulnerable to XXE attacks via malicious XML-RPC payloads. Use defusedxml.xmlrpc for protection.

Mako templates execute Python expressions without sandboxing. Ensure templates are from trusted sources and user input is not rendered as template code.

csv.writer() detected. Audit CSV output for formula injection (CSV injection) if data is exported to spreadsheet applications.

uuid.uuid1() embeds the host MAC address in the generated UUID, leaking hardware identity information.

os.chmod() or os.fchmod() sets overly permissive file permissions that allow unauthorized read, write, or execute access.

A function defines a default argument whose name suggests a password or secret but whose value is a hardcoded string literal.

Detects re.compile(), re.match(), re.search(), and re.findall() calls that should be audited for catastrophic backtracking patterns.

Detects logging.config.listen() which opens a socket that accepts and executes arbitrary logging configuration, enabling remote code execution.

Detects logging calls (info, debug, warning, error, critical) that should be audited for accidental credential or secret leakage in log output.