vLLM before 0.29.0 validates allowed_token_ids against tokenizer length instead of model output logits width in SamplingParams._validate_allowed_token_ids(). Attackers can supply token IDs above the output vocabulary that pass validation, causing LogitBiasState to corrupt GPU logits state and allow concurrent requests to sample tokens outside their allowlists.
GitPython 3.1.59 contains a vulnerability in its high-level diff API where the --no-index option bypasses protections against unsafe git diff options. This allows an attacker who can influence diff options and paths to use a blind local-file content oracle to infer contents of local files accessible to the GitPython process. The issue arises because --no-index treats path arguments as arbitrary filesystem paths, and when combined with the -I/--ignore-matching-lines option, it creates a content-dependent Boolean oracle. This vulnerability was confirmed in version 3.1.59 and is fixed in version 9.26.0 of the affected product, snakemake. Exploitation requires the embedding application to expose success/error distinctions and allow repeated queries with attacker-controlled inputs.
rustls-webpki through 0.103.12 (and 0.104.0-alpha releases before 0.104.0-alpha.7) contains a reachable panic in bit_string_flags() in src/der.rs. The input guard fails to reject a named-bit BIT STRING whose content is exactly [0x00] (zero padding bits and no data bytes), so raw_bits.len() - 1 underflows on the empty slice and the subsequent index operation panics (subtract-with-overflow in debug, index-out-of-bounds in release). The condition is reachable through the public API BorrowedCertRevocationList::from_der() when a CRL contains an issuingDistributionPoint extension with such an onlySomeReasons value. Exploitation requires an application that explicitly opts in to CRL revocation checking by passing RevocationOptions to verify_for_usage() and that parses CRL bytes obtained from a source the attacker can influence; the default rustls configuration, which does not use RevocationOptions, is unaffected. A crafted CRL causes a denial of service via the panic. Fixed in 0.103.13 and 0.104.0-alpha.7.
ArcadeDB (Maven artifact com.arcadedb:arcadedb-engine) through 26.8.1 enforces its per-type/per-record access-control rules only in LocalBucket, keyed on file id. Query-execution paths that reach record data through LSM index files or the TimeSeries engine never invoke that permission check, so an authenticated user who is denied readRecord/deleteRecord on a type can still, with a single ordinary SQL statement, read the type's indexed key values and record IDs (e.g. SELECT key, rid FROM INDEX:Type[field]), read MAX/MIN values via the index shortcut, read and count TimeSeries samples, learn the type's record count, and delete index entries (DELETE FROM INDEX:Type[field]), which desynchronizes the index from the data and can defeat unique constraints. Index and type names needed for exploitation are discoverable because SELECT FROM schema:indexes is unfiltered. The issue affects both embedded and server deployments and all transports (HTTP, Bolt, Postgres, Gremlin) once a principal is bound. Fixed in 26.9.1.
vLLM versions before 0.28.0 fail to validate the lower bound of token IDs in the /v1/embeddings and /pooling endpoints, allowing unauthenticated attackers to crash the engine by submitting negative token IDs. A single request with a negative token ID triggers a CUDA device-side assertion that poisons the GPU context, causing all subsequent requests to fail until the process restarts.
---
_-= Per source details. Do not edit below this line.=-_
## Source: amazon-inspector (826e81131e7b2a296324d469c10f1e44d5b33683703c8a2e0f101a8674c1cfcf)
The package impersonates the Tink open-banking SDK (name @tink/tink-link-core, description 'Enterprise-grade client core for open banking', homepage github.com/tink-link/core) but its actual behavior is host reconnaissance and DNS exfiltration. The package.json install script runs `node index.js`, and both the install entry and main entry load lib/core.js, which reads os.userInfo().username, os.hostname(), and process.cwd(), joins them with a prefix and a hardcoded domain as DNS labels, and issues a dns.resolve4() lookup. The destination hostname and module/API names are reconstructed at runtime from hex byte arrays via Buffer.from(...).toString('utf8') in lib/b02e30.js and lib/6ad264.js — the decoded domain is oob.algamil7x.xyz and the label prefix is tinkcore. Module names 'os','dns','process' and API names 'userInfo','hostname','cwd','resolve4' are loaded via module.constructor._load with hex-decoded strings to hide them from static review. DNS-based egress bypasses HTTP proxy and firewall controls. The beacon fires automatically on `npm install` (via the install lifecycle script) and again on `require()` of the package, giving the attacker installer username, hostname, and current working directory for every affected environment.
---
_-= Per source details. Do not edit below this line.=-_
## Source: amazon-inspector (c1ac938360c682e2489d11de9278697e7e2e85d73c612d2ab60ed28553bb5573)
The package's preinstall script (scripts.preinstall = 'node index.js') runs automatically on npm install and collects os.userInfo().username and os.hostname(), then issues a dns.lookup against a subdomain of 31ee29fe-db68-4fd3-86a2-2b707b9e95f0.dnshook.site with the username and hostname embedded in the queried label. This is a DNS-based exfiltration beacon: installing the package transmits installer host identifiers to a non-first-party canary domain over DNS resolution at install time. The version number 99.99.99 and package name shape are consistent with a dependency-confusion lure targeting an internal package name.
CubeCart is an ecommerce software solution. Prior to 6.7.5, the reset_id download-counter action and delete_card stored-payment-card action in admin/sources/orders.index.inc.php use state-changing GET requests and are omitted from the protection map in admin/skins/default/csrf.inc.php. A remote attacker can induce an authenticated administrator to issue one of these requests without a validated session token, causing unintended resets of electronic download usage counters or deletion of stored customer payment-card tokens. This issue is fixed in version 6.7.5.
A denial of service vulnerability was found in Immutable.js, a library used for persistent immutable data structures. The flaw occurs when certain List operations receive large index or size values, potentially causing an uncatchable infinite loop or unbounded memory allocation. This can disrupt applications using the library, including Red Hat products that process untrusted input through these List operations. Silent data corruption due to integer wrapping is also possible. Mitigation involves input validation to restrict large index values and isolating request handling processes with capped memory limits.
A vulnerability in the Linux kernel's AMD GPU display driver (drm/amd/display) involved an off-by-one error in handling the maximum number of connectors for MST HDCP arrays. The issue was due to a mismatch between the defined maximum display index and the actual maximum display count, causing array sizing errors. This vulnerability has been resolved by renaming and resizing the relevant constants to correctly reflect the intended maximum number of connectors.
A vulnerability in the Linux kernel's AMDGPU powerplay driver allows out-of-bounds indexing of voltage lookup tables due to missing bounds checks on voltage indices parsed from VBIOS tables. This flaw has been resolved by adding checks that return an error when indices are out of range.
A vulnerability in the Linux kernel's AMD GPU powerplay driver (drm/amdgpu/pm/powerplay) allows out-of-bounds indexing of voltage tables for Vega10 GPUs. The flaw involves missing bounds checks on voltage indices parsed from VBIOS tables, which can lead to invalid memory access. This issue has been resolved by adding bounds checks that return an error when indices are out of range.
A vulnerability in the Linux kernel's RDMA/hfi1 component was resolved. The issue involves improper handling of device allocation failures in hfi1_free_devdata(), which can lead to the erroneous removal of an unrelated device from the unit table at index 0. This occurs because the function assumes the device was inserted into the unit table and unconditionally erases dd->unit, but if allocation fails, the zero-initialized unit remains zero, causing cleanup to affect the wrong device.
A vulnerability in the Linux kernel's platform/x86 dell-wmi-base driver was fixed to correct the handling of the ultra performance key. The issue involved incorrect indexing when accessing event data fields, which could lead to a potential buffer overread. This flaw was addressed by adjusting the code to use the correct index for the first data field in the event data. The vulnerability affects multiple specific versions of the linux-hwe-edge kernel. No CVSS score is provided, but the impact is assessed as medium severity.
A vulnerability in the Linux kernel's HID driver for Roccat Kone devices allows a malicious USB device to cause an out-of-bounds read by supplying an invalid profile index. This occurs because the driver uses an 8-bit device-supplied profile value as an index into a 5-element array without proper range checking. The flaw was identified through static analysis and confirmed with a KUnit test. The issue has been resolved by rejecting out-of-range indices in the affected code paths.
A high-severity vulnerability in the Linux kernel's software node subsystem was fixed. The function software_node_get_reference_args() had an improper bounds check that failed when the index parameter was UINT_MAX, leading to an out-of-bounds access in the property array. This flaw could allow unauthorized access or modification of kernel memory. Multiple Linux kernel versions are affected, including various Ubuntu kernel builds. The issue has been resolved by correcting the bounds check to properly handle the UINT_MAX case.
A vulnerability in the Linux kernel's UBI subsystem was resolved that affected rollback handling during module initialization. The issue occurs because the rollback logic assumed the parameter index matched the UBI device number, which is not always true when an explicit ubi_num is specified. This could cause a failure to properly detach a successfully attached device during rollback. The fix involves scanning the entire ubi_devices array and detaching devices by their actual array index.
A vulnerability in the Linux kernel's mt76 wifi driver for the mt7915 device caused a double initialization of the hif2 interface on the non-WED path. This double initialization led to incrementing a global index twice, writing to a PCIe register twice, and leaking a device reference. The issue was fixed by removing the unconditional call to the initialization function, leaving only the correct call inside the WED-inactive branch.