google/security-research

Arbitrary file read in the model loading mechanism (HDF5 integration) in Keras versions 3.0.0 through 3.13.1 on all supported platforms allows a remote attacker to read local files and disclose sensitive information via a crafted .keras model file utilizing HDF5 external dataset references.

Entrust nShield Connect XC, nShield 5c, and nShield HSMi through 13.6.11, or 13.7, allow a Physically Proximate Attacker to access the internal components of the appliance, without leaving tamper evidence. To exploit this, the attacker needs to remove the tamper label and all fixing screws from the device without damaging it. This is called an F14 attack.

Entrust nShield Connect XC, nShield 5c, and nShield HSMi through 13.6.11, or 13.7, allow an attacker to gain access the the BIOS menu because is has no password.

Entrust nShield Connect XC, nShield 5c, and nShield HSMi through 13.6.11, or 13.7, allow a Physically Proximate Attacker to Escalate Privileges by enabling the USB interface through chassis probe insertion during system boot, aka "Unauthorized Reactivation of the USB interface" or F01.

Entrust nShield Connect XC, nShield 5c, and nShield HSMi through 13.6.11, or 13.7, might allow a physically proximate attacker to gain access to the EOL legacy bootloader.

Entrust nShield Connect XC, nShield 5c, and nShield HSMi through 13.6.11, or 13.7, allow a physically proximate attacker to escalate privileges by booting from a USB device with a valid root filesystem. This occurs because of insecure default settings in the Legacy GRUB Bootloader.

Entrust nShield Connect XC, nShield 5c, and nShield HSMi through 13.6.11, or 13.7, allow a user with OS root access to alter firmware on the Chassis Management Board (without Authentication). This is called F04.

Entrust nShield Connect XC, nShield 5c, and nShield HSMi through 13.6.11, or 13.7, allow a physically proximate attacker to modify or erase tamper events via the Chassis management board.

Entrust nShield Connect XC, nShield 5c, and nShield HSMi through 13.6.11, or 13.7, allow a physically proximate attacker with elevated privileges to falsify tamper events by accessing internal components.

Entrust nShield Connect XC, nShield 5c, and nShield HSMi through 13.6.11, or 13.7, allow a physically proximate attacker (with elevated privileges) to read and modify the Appliance SSD contents (because they are unencrypted).

Entrust nShield Connect XC, nShield 5c, and nShield HSMi through 13.6.11, or 13.7, allow a physically proximate attacker with root access to modify the Recovery Partition (because of a lack of integrity protection).

Entrust nShield Connect XC, nShield 5c, and nShield HSMi through 13.6.11, or 13.7, allow a physically proximate attacker to escalate privileges by editing the Legacy GRUB bootloader configuration to start a root shell upon boot of the host OS. This is called F06.

The Chassis Management Board in Entrust nShield Connect XC, nShield 5c, and nShield HSMi through 13.6.11, or 13.7, allows a physically proximate attacker to persistently modify firmware and influence the (insecurely configured) appliance boot process. To exploit this, the attacker must modify the firmware via JTAG or perform an upgrade to the chassis management board firmware. This is called F03.

The Chassis Management Board in Entrust nShield Connect XC, nShield 5c, and nShield HSMi through 13.6.11, or 13.7, allows a physically proximate attacker to obtain debug access and escalate privileges by bypassing the tamper label and opening the chassis without leaving evidence, and accessing the JTAG connector. This is called F02.

A heap-buffer-overflow write exists in jpeg2000dec FFmpeg which allows an attacker to potentially gain remote code execution or cause denial of service via the channel definition cdef atom of JPEG2000.

An integer overflow exists in the FTS5 https://sqlite.org/fts5.html  extension. It occurs when the size of an array of tombstone pointers is calculated and truncated into a 32-bit integer. A pointer to partially controlled data can then be written out of bounds.

Fullscreen API Spoofing and UI Redressing in the handling of Fullscreen API and UI rendering in OpenAI Operator SaaS on Web allows a remote attacker to capture sensitive user input (e.g., login credentials, email addresses) via displaying a deceptive fullscreen interface with overlaid fake browser controls and a distracting element (like a cookie consent screen) to obscure fullscreen notifications, tricking the user into interacting with the malicious site.

There is a XXE in W3CSS Validator versions before cssval-20250226 that allows an attacker to use specially-crafted XML objects to coerce server-side request forgery (SSRF).  This could be exploited to read arbitrary local files if an attacker has access to exception messages.

There exists an unauthenticated accessible JTAG port on the Kioxia PM6, PM7 and CM6 devices - On the Kioxia CM6, PM6 and PM7 disk drives it was discovered that the 2 main CPU cores of the SoC can be accessed via an open JTAG debug port that is exposed on the drive’s circuit board. Due to the wide cutout of the enclosures, the JTAG port can be accessed without having to open the disk enclosure. Utilizing the JTAG debug port, an attacker with (temporary) physical access can get full access to the firmware and memory on the 2 main CPU cores within the drive including the execution of arbitrary code, the modification of firmware execution flow and data or bypassing the firmware signature verification during boot-up.

Multiple stored cross-site scripting (“XSS”) vulnerabilities in the markdown dashboard and dashboard comment functionality of Lightdash version 0.1024.6 allows remote authenticated threat actors to inject malicious scripts into vulnerable web pages. A threat actor could potentially exploit this vulnerability to store malicious JavaScript which executes in the context of a user’s session with the application.

Lightdash version 0.1024.6 allows users with the necessary permissions, such as Administrator or Editor, to create and share dashboards. A dashboard that contains HTML elements which point to a threat actor controlled source can trigger an SSRF request when exported, via a POST request to /api/v1/dashboards//export. The forged request contains the value of the exporting user’s session token. A threat actor could obtain the session token of any user who exports the dashboard. The obtained session token can be used to perform actions as the victim on the application, resulting in session takeover.

A boolean-based SQL injection issue in the Virtual Meeting Password (VMP) endpoint in R-HUB TurboMeeting through 8.x allows unauthenticated remote attackers to extract hashed passwords from the database, and authenticate to the application, via crafted SQL input.

A command-injection issue in the Certificate Signing Request (CSR) functionality in R-HUB TurboMeeting through 8.x allows authenticated attackers with administrator privileges to execute arbitrary commands on the underlying server as root.

The password-reset mechanism in the Forgot Password functionality in R-HUB TurboMeeting through 8.x allows unauthenticated remote attackers to force the application into resetting the administrator's password to a random insecure 8-digit value.

A user who can create objects in a database with plv8 3.2.1 installed is able to cause deferred triggers to execute as the Superuser during autovacuum.

JPX Fragment List (flst) box vulnerability in Kakadu 7.9 allows an attacker to exfiltrate local and remote files reachable by a server if the server allows the attacker to upload a specially-crafted the image that is displayed back to the attacker.

An issue was discovered in net/ceph/messenger_v2.c in the Linux kernel before 6.4.5. There is an integer signedness error, leading to a buffer overflow and remote code execution via HELLO or one of the AUTH frames. This occurs because of an untrusted length taken from a TCP packet in ceph_decode_32.

The Linux kernel allows userspace processes to enable mitigations by calling prctl with PR_SET_SPECULATION_CTRL which disables the speculation feature as well as by using seccomp. We had noticed that on VMs of at least one major cloud provider, the kernel still left the victim process exposed to attacks in some cases even after enabling the spectre-BTI mitigation with prctl. The same behavior can be observed on a bare-metal machine when forcing the mitigation to IBRS on boot command line. This happened because when plain IBRS was enabled (not enhanced IBRS), the kernel had some logic that determined that STIBP was not needed. The IBRS bit implicitly protects against cross-thread branch target injection. However, with legacy IBRS, the IBRS bit was cleared on returning to userspace, due to performance reasons, which disabled the implicit STIBP and left userspace threads vulnerable to cross-thread branch target injection against which STIBP protects.

In OpenBMC 2.9, crafted IPMI messages allow an attacker to cause a denial of service to the BMC via the netipmid (IPMI lan+) interface.

In Epiphany (aka GNOME Web) through 43.0, untrusted web content can trick users into exfiltrating passwords, because autofill occurs in sandboxed contexts.

SnakeYaml's Constructor() class does not restrict types which can be instantiated during deserialization. Deserializing yaml content provided by an attacker can lead to remote code execution. We recommend using SnakeYaml's SafeConsturctor when parsing untrusted content to restrict deserialization. We recommend upgrading to version 2.0 and beyond.

Dm-verity is used for extending root-of-trust to root filesystems. LoadPin builds on this property to restrict module/firmware loads to just the trusted root filesystem. Device-mapper table reloads currently allow users with root privileges to switch out the target with an equivalent dm-linear target and bypass verification till reboot. This allows root to bypass LoadPin and can be used to load untrusted and unverified kernel modules and firmware, which implies arbitrary kernel execution and persistence for peripherals that do not verify firmware updates. We recommend upgrading past commit 4caae58406f8ceb741603eee460d79bacca9b1b5

In OpenBMC 2.9, crafted IPMI messages allow an attacker to bypass authentication and gain full control of the system.

A heap out-of-bounds write affecting Linux since v2.6.19-rc1 was discovered in net/netfilter/x_tables.c. This allows an attacker to gain privileges or cause a DoS (via heap memory corruption) through user name space

An issue was discovered in Linux: KVM through Improper handling of VM_IO|VM_PFNMAP vmas in KVM can bypass RO checks and can lead to pages being freed while still accessible by the VMM and guest. This allows users with the ability to start and control a VM to read/write random pages of memory and can result in local privilege escalation.

A weak robustness vulnerability exists in the AWS Encryption SDKs for Java, Python, C and Javalcript prior to versions 2.0.0. Due to the non-committing property of AES-GCM (and other AEAD ciphers such as AES-GCM-SIV or (X)ChaCha20Poly1305) used by the SDKs to encrypt messages, an attacker can craft a unique cyphertext which will decrypt to multiple different results, and becomes especially relevant in a multi-recipient setting. We recommend users update their SDK to 2.0.0 or later.

Telestream Tektronix Medius before 10.7.5 and Sentry before 10.7.5 have a SQL injection vulnerability allowing an unauthenticated attacker to dump database contents via the page parameter in a page=login request to index.php (aka the server login page).

A vulnerability in the in-band key negotiation exists in the AWS S3 Crypto SDK for GoLang versions prior to V2. An attacker with write access to the targeted bucket can change the encryption algorithm of an object in the bucket, which can then allow them to change AES-GCM to AES-CTR. Using this in combination with a decryption oracle can reveal the authentication key used by AES-GCM as decrypting the GMAC tag leaves the authentication key recoverable as an algebraic equation. It is recommended to update your SDK to V2 or later, and re-encrypt your files.

A padding oracle vulnerability exists in the AWS S3 Crypto SDK for GoLang versions prior to V2. The SDK allows users to encrypt files with AES-CBC without computing a Message Authentication Code (MAC), which then allows an attacker who has write access to the target's S3 bucket and can observe whether or not an endpoint with access to the key can decrypt a file, they can reconstruct the plaintext with (on average) 128*length (plaintext) queries to the endpoint, by exploiting CBC's ability to manipulate the bytes of the next block and PKCS5 padding errors. It is recommended to update your SDK to V2 or later, and re-encrypt your files.