An incident involving the Qwen3.8-Flash-Next AI model from Alibaba showed the model generating a shell command that attempted to send data to an Alibaba Cloud storage address. Investigation revealed that the model's behavior was inconsistent, producing a wide variety of destination URLs including many local, nonexistent, or invalid domains. This variability suggests the behavior was due to model jaggedness—unpredictable changes in output based on small context differences—rather than a fixed backdoor. The model's reasoning contradicted its action, and changes in input ordering affected the likelihood of network calls. The incident highlights the importance of sandboxing and least privilege controls when deploying AI agents with tool access.
CVE-2026-93010 is a vulnerability in dnsmasq related to the DHCP leasequery feature. When leasequery is enabled without restricting allowed source addresses, the server does not verify the origin of lease queries. This allows any host on the local network to request DHCP lease records, revealing assigned IP addresses, MAC addresses, hostnames, and lease metadata. The feature is disabled by default and must be explicitly enabled by an administrator.
CVE-2026-93006 is a vulnerability in dnsmasq's TFTP server where a transfer cache is not invalidated after a transfer finishes. This flaw can cause a client to receive previously cached data instead of the requested file, enabling an attacker on the local network to substitute attacker-chosen boot content in PXE network-boot environments. The vulnerability has a high impact on confidentiality and integrity but requires high attack complexity and local network access.
CVE-2026-17591 is a use-after-free vulnerability in libarchive affecting versions prior to 3.7.1-14.oe2403sp1. The flaw occurs in the archive_entry_copy_mac_metadata() function, where an internal buffer is deallocated before the data copy completes. This can lead to a Denial of Service (application crash) or potential information disclosure by reading uninitialized memory. The vulnerability has a low CVSS score of 3.6 and requires local attacker interaction with high attack complexity. A patch is available in version 3.7.1-14.oe2403sp1.
Multiple critical security vulnerabilities have been identified in Apache Tomcat versions ranging from early milestone releases through recent stable versions up to 11.0.26, 10.1.60, and 9.0.122. These include authentication bypass, request/response mix-up in HTTP/2 asynchronous servlets, failure to securely enforce FIPS mode, execution after redirect, local unauthorized access via a TOCTOU race condition in unix domain sockets, and incomplete cleanup issues causing information leaks and TLS 1.3 connection failures. Users are strongly recommended to upgrade to fixed versions 11.0.27, 10.1.61, or 9.0.123 to address these vulnerabilities.
Rclone versions before 1.75.2 have a path traversal vulnerability in the local backend's translateLink function. This flaw occurs because the function trims the .rclonelink suffix after performing a containment check, allowing a path to resolve outside the intended destination root. An attacker can exploit this by planting a specially named directory in a source tree synced with --links and --metadata, potentially modifying ownership, timestamps, and extended attributes of the destination's parent directory.
Rclone versions from 1.52.0 up to but not including 1.75.2 contain a cross-site request forgery (CSRF) vulnerability in the pCloud backend's localhost OAuth callback. This flaw disables the state check and trusts the hostname parameter, allowing malicious web pages to send unauthorized requests to the local OAuth callback port. As a result, attacker tokens and API hosts can be planted, causing subsequent transfers to be redirected to the attacker.
A vulnerability in the Node.js zlib.ZipFile API allows it to return bytes not belonging to the requested ZIP entry when local and central header validation fails. This occurs because the parsed local header is cached without a content offset, causing subsequent reads to return attacker-controlled bytes from the start of the file. The CRC-32 check still passes due to attacker control of the central directory CRC. This can cause Node.js-based ZIP scanners to approve different content than other ZIP implementations extract. The flaw affects Node.js versions 25.0.0 up to but not including 26.12.1 and is fixed in 26.12.1.
A critical vulnerability in the zip crate (<= 9.0.4) arises from inconsistent parsing of ZIP archive headers between two readers: ZipArchive uses the Central Directory File Header (CDFH) as authoritative, while ZipStreamReader relies on the Local File Header (LFH). This discrepancy allows an attacker to craft a ZIP archive with differing filenames in the headers, enabling security gateways to be bypassed and malicious files to be extracted, leading to arbitrary file overwrite and potential remote code execution.
A race condition vulnerability in filelock versions 3.30.0 through 4.2.0 allows a local unprivileged user to swap a symbolic link at the <path>.filelock directory after an lstat check but before subsequent operations. This can cause the lock user's process to create claim files in an attacker-chosen directory, potentially disrupting other locks by causing stranded claims that lead to denial of service until manually force-broken. The vulnerability requires write access to the directory containing <path>.filelock and is mitigated by patch 4.2.1.
MySqlConnector versions prior to 2.6.1 have a vulnerability in zero-configuration TLS mode when using SslMode set to VerifyCA or VerifyFull without a configured CA certificate. In this scenario, an attacker capable of man-in-the-middle interception can present a self-signed certificate and cause the client to send its database password in clear text before the certificate is validated. This undermines the expected server authentication and exposes the password to interception. The issue does not affect connections that use a configured CA certificate, local Unix socket connections, or empty passwords. A fix is available in version 2.6.1.
CVE-2026-61542 is a high-severity vulnerability in drhelius Gearboy emulator version prior to 3.8.7. The Gearsystem component's optional --mcp-http mode accepts unauthenticated MCP requests from any network peer, allowing remote peers to invoke emulator and debugging operations without authentication. The vulnerability arises because the MCP server binds to all interfaces and does not enforce authorization checks on POST requests to /mcp. The impact is limited to the emulator process, debugging state, and local files used as media, with no demonstrated host code execution. Gearboy 3.8.7 includes a patch that fixes this issue.
CVE-2026-106484 is a medium severity vulnerability in Google Chrome's chrome-devtools-mcp component on Linux. It involves insufficient policy enforcement allowing a local attacker to perform arbitrary read and write operations via a malicious file. This issue affects versions from 1.1.0 up to but not including 1.5.0 and was fixed in version 1.5.0.
CVE-2026-103404 is a vulnerability in Juniper Networks Session Smart Conductor's RoutingManager component where incorrect permission assignments allow a local user to access sensitive routing credentials. This occurs through triggering processes that write to the /tmp directory. The issue affects multiple versions prior to fixed releases 7.0.5, 7.1.7, and 7.2.3. The vulnerability has a CVSS score of 6.8, indicating a medium severity level.
A vulnerability in the Linux kernel's io_uring subsystem causes an exiting task to spin indefinitely on local work it cannot run, leading to CPU resource exhaustion. This occurs when a disabled io_uring ring is enabled by another task, and the exiting task has pending requests it cannot complete. The issue can cause a kernel panic if panic_on_warn is enabled.
The Bit Flows plugin for WordPress up to version 1.31.0 is vulnerable to a Local File Inclusion (LFI) flaw via the 'image' parameter used by the ImageHelper node. This vulnerability allows unauthenticated attackers to include and execute arbitrary PHP files on the server if certain webhook configurations are in place, potentially leading to code execution or access control bypass. Exploitation requires an administrator to have configured a webhook-triggered flow using the vulnerable node without additional security controls like bearer secrets or IP allowlists.