PraisonAI versions before 4.5.128 have a vulnerability allowing authenticated users to execute arbitrary shell commands. This occurs because the UI modules hardcode the approval_mode to auto, ignoring the administrator's configuration set via the PRAISON_APPROVAL_MODE environment variable. The flaw enables attackers to bypass manual approval and command sanitization controls.
CVE-2025-15661 is an out-of-bounds heap read vulnerability in libssh2 versions up to 1.11.1. The flaw exists in the sftp_symlink() function, where a malicious SSH server or man-in-the-middle attacker can send a crafted SSH_FXP_NAME response with an oversized link_len value. This triggers a heap buffer over-read, potentially disclosing heap memory contents or causing a crash. The issue arises from missing validation of the packet buffer size before a memcpy operation. The vulnerability has a high severity score of 8.3 and was fixed in a commit identified as 2dae302.
Typemill versions prior to 2.24.0 have a path traversal vulnerability that allows authenticated users with Author-level privileges to read arbitrary files outside the intended content directory. This occurs due to insufficient restriction on pathname inputs in the Storage::getFile() function, enabling bypass of traversal-prevention controls. The vulnerability is fixed in version 2.24.0.
CVE-2026-54388 is a critical vulnerability in tinyproxy versions up to 1.11.3 involving HTTP request/response smuggling. The proxy fails to reject HTTP requests containing multiple Content-Length headers with differing values, forwarding all duplicate headers to the backend while using only the first header's value to determine the request body length. This inconsistency allows remote attackers to desynchronize the proxy and backend parser states, enabling injection of arbitrary HTTP requests to the backend. Exploitation can lead to cache poisoning, access control bypass, and request hijacking. The issue is fixed in a commit identified as 364cdb6.
CVE-2026-54387 is a critical vulnerability in tinyproxy through version 1.11.3 where the proxy inconsistently handles HTTP requests containing both Content-Length and Transfer-Encoding: chunked headers. This flaw allows remote attackers to desynchronize the proxy and backend server's HTTP parser states, enabling injection of arbitrary HTTP requests to the backend. The vulnerability can lead to cache poisoning, access control bypass, and request hijacking. It was fixed in commit ff45d3b.
CVE-2026-55199 is a denial of service vulnerability in libssh2 up to version 1.11.1. It occurs in the SSH_MSG_EXT_INFO handler during key exchange, where a malicious SSH server can send a crafted extension count value (nr_extensions set to 0xFFFFFFFF) causing the client to enter a CPU exhaustion loop for over 60 seconds. This happens because return values from _libssh2_get_string() are unchecked and session timeouts do not apply to CPU-bound loops. The issue was fixed in a commit identified as 1762685.
CVE-2026-53875 is a high-severity vulnerability in picklescan before version 1.0.3. It involves improper neutralization of directives in dynamically evaluated code, specifically an 'eval injection' in the scan_pytorch function. This flaw allows attackers to embed malicious magic numbers using the __reduce__ trick, enabling crafted PyTorch payloads to bypass detection while remaining executable. Successful exploitation leads to arbitrary code execution when the malicious payload is loaded with torch.load().
picklescan versions before 1.0.1 contain an unsafe deserialization vulnerability that allows unauthenticated attackers to execute arbitrary code by embedding malicious eval calls within pickle files. This vulnerability arises from the ability to hide eval calls nested under callable objects via getattr, enabling code execution when loading pickle data from untrusted sources.
picklescan versions before 1.0.4 have an incomplete blocklist in the profile module that does not block the profile.run() function. This allows attackers to craft malicious pickle files that execute arbitrary Python code via exec(), bypassing picklescan's security checks.
picklescan versions before 0.0.35 contain an unsafe pickle deserialization vulnerability that allows unauthenticated attackers to read arbitrary server files. This is achieved by chaining io.FileIO and urllib.request.urlopen, enabling attackers to bypass remote code execution blocklists and exfiltrate sensitive data such as /etc/passwd to external servers. The vulnerability is fixed in version 0.0.35.
CVE-2025-71325 is a critical vulnerability in picklescan versions before 0.0.27. It involves a parsing logic error in the _list_globals function when processing STACK_GLOBAL opcodes, which causes improper tracking of arguments. This flaw allows crafted malicious pickle files with arguments at position zero to bypass security scanning by triggering unexpected exceptions.
CVE-2025-71323 is a critical vulnerability in picklescan versions before 0.0.33 where the software fails to block the ctypes module. This allows attackers to craft malicious pickle files that leverage ctypes.WinDLL to load kernel32.dll and execute arbitrary commands, bypassing sandbox protections and gadget chain detection. The vulnerability enables remote code execution without requiring user interaction or privileges.
PickleScan versions before 0.0.33 have a security flaw where the pty.spawn function is not included in the unsafe globals list. This omission allows attackers to craft malicious pickle payloads that can bypass security checks and execute arbitrary code when processed by vulnerable PickleScan versions.
picklescan versions before 0.0.33 contain a critical deserialization vulnerability that allows attackers to bypass security blocklists and write arbitrary files. This flaw enables malicious pickle objects to overwrite important system files, potentially causing denial of service or remote code execution.
CVE-2025-71320 is a critical vulnerability in picklescan versions before 0.0.33. The software's deny-list for disallowed inputs is incomplete, failing to block the pydoc.locate and operator.methodcaller functions. This allows remote attackers to craft malicious pickle files that bypass security checks and achieve arbitrary code execution upon deserialization. The issue is fixed in version 0.0.33.
CVE-2026-53868 is a denial of service vulnerability in Capgo versions before 12.128.2. It allows attackers to register accounts with arbitrary email addresses without verification and then initiate deletion of those accounts. This causes the targeted email addresses to be locked in a pending deletion state, preventing legitimate users from accessing the platform for 30 days.
Cap-go versions prior to 12.128.2 have an account takeover vulnerability in the email change process. An attacker with a temporary authenticated session can change the registered email address without requiring re-authentication such as password or MFA verification. This allows the attacker to redirect verification emails to an attacker-controlled address and then reset the victim's password, resulting in permanent account takeover.
CVE-2025-71329 is a denial of service vulnerability in the image-size library up to version 2.0.2. It allows remote attackers to cause an infinite loop in the Node.js event loop by supplying a specially crafted image buffer with a zero-valued size field in a recognized box-type. This triggers an infinite loop in the JXL or HEIF image parsers, causing the application to hang permanently.
CVE-2025-71330 is a denial of service vulnerability in the image-size library up to version 2.0.2. It allows remote attackers to cause an infinite loop in the ICNS image parser by supplying a specially crafted ICNS buffer with a zero-valued entry length field. This infinite loop blocks the Node.js event loop permanently, causing a denial of service.
CVE-2025-71319 is a denial of service vulnerability in the image-size library versions 1.1.0 through 1.2.1 and 2.0.0 through 2.0.2. It occurs when processing specially crafted JXL, HEIF, or JP2 image files containing zero-sized boxes, causing an infinite loop in the findBox function. This infinite loop results in the Node.js event loop being permanently blocked, hanging the application and disrupting service availability.