| CVE |
Vendors |
Products |
Updated |
CVSS v3.1 |
| llama.cpp builds b3978 through b9058 contain an integer underflow and out-of-bounds read vulnerability in the DRY sampler that allows unauthenticated attackers to trigger a heap buffer underflow by sending a crafted HTTP request with dry_allowed_length set to INT32_MIN to the /v1/completions or /v1/chat/completions endpoints. Attackers can exploit this vulnerability to crash the server with SIGSEGV causing denial of service for all connected users, or corrupt token sampling probabilities by reading garbage values from memory before the allocated buffer. |
| llama.cpp builds b1886 through b7445 contain an integer overflow vulnerability in the LLaMA-Android JNI wrapper where the new_1batch() function multiplies sizeof(llama_seq_id) by an attacker-controlled n_seq_max parameter without overflow validation, causing heap buffer allocation to wrap and allocate insufficient memory. Attackers can exploit this by providing a crafted n_seq_max value through a malicious model file or JNI call to trigger heap corruption and achieve denial of service or arbitrary code execution on Android applications using the LLaMA-Android binding. |
| A regression exists in the Linux Kernel within KVM: nVMX that allowed for speculative execution attacks. L2 can carry out Spectre v2 attacks on L1 due to L1 thinking it doesn't need retpolines or IBPB after running L2 due to KVM (L0) advertising eIBRS support to L1. An attacker at L2 with code execution can execute code on an indirect branch on the host machine. We recommend upgrading to Kernel 6.2 or applying the relevant stable backports (v5.4.233, v5.10.170, v5.15.96, v6.1.14). |
| A maliciously crafted TIF file, when parsed through certain Autodesk products during image import, can cause an Out-of-Bounds Read in the image handling library. A malicious actor can leverage this vulnerability to cause a denial of service |
| A maliciously crafted DWG or DXF file, when parsed through Autodesk AutoCAD, can force an Out-of-Bounds Read vulnerability. A malicious actor can leverage this vulnerability to cause a crash or disclose sensitive information. |
| A maliciously crafted DWG or DXF file, when parsed through Autodesk AutoCAD, can force an Out-of-Bounds Read vulnerability. A malicious actor can leverage this vulnerability to cause a crash or disclose sensitive information. |
| A maliciously crafted DXF file, when parsed through Autodesk AutoCAD, can force a Heap-Based Overflow vulnerability. A malicious actor can leverage this vulnerability to cause a crash, read sensitive data, or execute arbitrary code in the context of the current process. |
| Contiki-NG's DNS/mDNS resolver skip_name() in os/services/resolv/resolv.c walks DNS wire-format name labels with no packet-boundary check, and the caller in newdata() invokes it in a loop iterating nquestions times from the attacker-controlled DNS header before validating the transaction ID. An attacker who sets nquestions higher than the number of complete questions present causes skip_name() to walk past the UDP packet buffer, and the returned pointer is cast to struct dns_answer * for further memory reads. On builds with RESOLV_CONF_SUPPORTS_MDNS enabled, any peer on the local segment can trigger the read unauthenticated via a multicast UDP 5353 packet with no outstanding query required; on standard DNS builds an attacker who can inject a UDP response from port 53 during an outstanding query can trigger the same read. Impact is out-of-bounds read of uip_buf and adjacent memory, disclosing memory contents or crashing the resolver. |
| Mermaid is a JavaScript tool that uses Markdown-inspired text to create and modify diagrams and charts. From version 11.5.0 until 11.16.1, Mermaid Architecture Diagrams are vulnerable to prototype pollution when a diagram defines a group with an id of __proto__. Because the group id is used directly as an object property key without validation, an attacker who can supply diagram text can pollute Object.prototype, potentially affecting the behavior of the embedding application. This issue is fixed in version 11.16.1. |
| PILOS (Platform for Interactive Live-Online Seminars) is a frontend for BigBlueButton. From 2.1.0 until 4.14.1, PILOS does not send a Cross-Origin-Opener-Policy response header, so pages opened by PILOS via a link that opens a new browsing context (e.g., target="_blank") retain a window.opener reference back to the originating PILOS tab. A malicious destination page reached this way can use window.opener to navigate or manipulate the original PILOS tab, a technique known as reverse tabnabbing, potentially redirecting an authenticated user to a phishing page that mimics PILOS. This issue is fixed in version 4.14.1. |
| league/commonmark is a PHP library for parsing and rendering CommonMark Markdown. From 0.6.0 until 2.9.0, specially crafted Markdown lines can cause the parser to have quadratic time complexity when converting, because several parsing paths repeatedly rescan growing portions of a line to translate between character positions and byte positions, and the Autolink extension can also copy and validate the remaining line at every URL-like prefix, allowing an attacker who can submit Markdown for conversion to consume disproportionate CPU time with a comparatively small request. This issue is fixed in 2.9.0. |
| TimescaleDB through 2.29.1, fixed in commit 517c13e, contains an out-of-bounds read vulnerability that allows authenticated attackers to cause query-result integrity failures or backend crashes by supplying a crafted Simple8b selector-11 value, which is stored in the signed int16 Arrow dictionary-index type and bypasses index validation checks in bulk text dictionary decompression. Attackers with direct DML access to a non-frozen physical compressed hypertable relation can trigger an out-of-bounds read before the base of the live offsets array through the VectorAgg single-text hashing strategy, resulting in incorrect aggregation output, backend SIGSEGV, or PostgreSQL crash recovery depending on build configuration. |
| TimescaleDB through 2.29.1, fixed in commit 517c13e, contains an out-of-bounds read in the Dictionary compression reverse row iterator (tsl/src/compression/algorithms/dictionary.c). The forward path validates the decoded index; the reverse path uses an assertion compiled out of release builds, leaving the 64-bit Simple8b index unvalidated and the read offset attacker-controlled. Attackers with DML access to a physical compressed relation can store a crafted datum and run a reverse-order scan. With a pass-by-value column type the out-of-bounds Datum is returned to the client as a normal column value, disclosing backend memory including the shared buffer pool, which SQL access control does not cover. |
| node-re2 provides RE2 regular expression bindings for Node.js. Prior to version 1.25.1, the WrappedRE2::Replace function built its replacement result and passed it to V8 using ToLocalChecked without checking for the empty MaybeLocal that V8 returns when the resulting string or buffer exceeds V8's maximum string length. When a global replace uses an output amplifying replacement template, the result can grow quadratically with the input size, and once the result exceeds V8's maximum string length, the unchecked ToLocalChecked call causes a fatal, uncatchable process abort instead of a catchable exception. This issue is fixed in version 1.25.1. |
| TimescaleDB through 2.29.1, fixed in commit 517c13e, contains an out-of-bounds read vulnerability in the Gorilla compression reverse row iterator that allows authenticated attackers to cause a denial of service by storing a crafted compressed datum with an internally inconsistent BitArray. Attackers with DML access to a compressed hypertable can trigger an unsigned integer wraparound in the reverse iterator bucket index computation, causing a read beyond the end of the bucket array, resulting in a SIGSEGV crash that can be repeatedly triggered on each subsequent reverse-order scan. |
| FFmpeg versions from 0.5 up to, but not including, 9.0 contain a signed integer overflow vulnerability in the DVB subtitle parser in libavcodec/dvbsub_parser.c that allows attackers to trigger a heap buffer overflow by supplying a crafted WTV file. The overflow causes the bounds-check guard expression to wrap to INT_MIN, bypassing the PARSE_BUF_SIZE comparison and invoking memcpy() with attacker-controlled data into a heap buffer, resulting in an out-of-bounds heap write and potential memory corruption or code execution. |
| GitHub CLI (gh) is GitHub's official command line tool. Prior to version 2.97.0, multiple GitHub CLI commands printed externally controlled gist, API, pull request, release, codespace, skill, or agent-task content without neutralizing terminal escape sequences. An attacker who can influence that content can embed escape sequences that are interpreted by the terminal of a user who runs an affected command, with impact ranging from cosmetic manipulation of the title or on-screen content to, on some terminal emulators, command execution. This extends the same class of issue as CVE-2026-45803—which addressed only gh run view --log—to the other affected command paths. This issue is fixed in version 2.97.0. |
| Mermaid is a JavaScript tool that uses Markdown-inspired text to create and modify diagrams and charts. From version 11.6.0 until 11.16.1, Mermaid Radar Diagrams allow arbitrary large values for the ticks parameter, which can cause high CPU usage and freeze the rendering webpage or JavaScript process for long periods of time, potentially until the process is killed from memory exhaustion. This issue is fixed in version 11.16.1. |
| node-re2 provides RE2 regular expression bindings for Node.js. Prior to version 1.26.1, passing a Buffer whose final bytes form a truncated (incomplete) multi-byte UTF-8 sequence could cause the native binding to read past the end of the allocated buffer while attempting to decode the final, incomplete code point. This could result in an out-of-bounds read and potential disclosure of adjacent memory contents. This issue is fixed in version 1.26.1. |
| pymdown-extensions is a collection of extensions for the Python Markdown library. In versions up to and including 11.0, four inline processors (caret, tilde, betterem, and magiclink) use regular expressions whose content groups can partition a run of delimiter characters in exponentially many ways, causing catastrophic backtracking. As a result, a single untrusted Markdown line under 50 bytes rendered with markdown.markdown() in each extension's default configuration drives the rendering thread into unbounded CPU usage that grows exponentially with input length, enabling an unauthenticated remote attacker who can submit Markdown to cause denial of service. The exposure is concrete for web applications that render user-supplied Markdown (comments, wikis, issue bodies, live preview), including any app using pymdownx.extra which bundles the vulnerable betterem default, as well as hosted docs/CI systems that build untrusted Markdown. The issue has been fixed in version 11.0.1. |
