llama.cpp builds b4882 through b9058 contain a heap buffer overflow vulnerability in the KV cache state restore path where the state_read_data() function computes write size without overflow checking, allowing attackers with write access to the slot_save_path directory to corrupt heap memory. Attackers can craft malicious state files where cell_count multiplication overflows or exceeds tensor buffer allocation to write attacker-controlled bytes past buffer boundaries, potentially resulting in heap metadata corruption, model weight corruption, or arbitrary code execution via function pointer overwrite.

Sonic 3 A.I.R. before commit 2492d18 contains an unbounded memory allocation vulnerability in ReceivedPacketCache::enqueuePacket() that allows unauthenticated remote attackers to crash the server process by sending a crafted UDP packet with mUniquePacketID set to the maximum uint32 value. The mUniquePacketID field is read directly from the UDP wire-format packet header without bounds checking, causing the server to allocate one CacheItem per missing packet ID gap, exhausting available host memory and propagating an uncaught std::bad_alloc exception to std::terminate().

When an Event Publisher output adapter is configured with irrelevant properties, the affected products log these properties. This logging occurs without sufficient validation or sanitization of the property values. A malicious actor with access to the 'wso2carbon' log files could retrieve sensitive information, such as user credentials or other confidential data, that was inadvertently logged due to misconfiguration, potentially leading to unauthorized access.

In Apache CXF's DefaultEncryptingCodeDataProvider, a captured authorization code can be redeemed an unlimited number of times due to a flaw in the implementation of the removeCodeGrant functionality. This violates the RFC requirement that "The authorization code MUST NOT be used more than once." Users are recommended to upgrade to versions 4.2.3 or 4.1.8 or 3.6.12, which fix this issue.

A vulnerability was determined in FoundationAgents MetaGPT up to 0.8.2. This affects the function read of the file metagpt/tools/libs/editor.py. This manipulation causes path traversal. The attack needs to be launched locally. The exploit has been publicly disclosed and may be utilized. The vendor was contacted early about this disclosure but did not respond in any way.

OpenReception's appointment booking software provides an end-to-end encrypted appointment booking platform. Prior to version 1.0.2, a TENANT_ADMIN promotes themselves to platform-wide GLOBAL_ADMIN through a single PUT request. The role-update handler accepts the `GLOBAL_ADMIN` enum value from any tenant admin updating their own tenant's staff. No policy check enforces that "only an existing GLOBAL_ADMIN may grant GLOBAL_ADMIN", so the schema validation IS the authorization decision. After re-login, the JWT contains the new role and the formerly-tenant-scoped admin reaches every other tenant on the platform. On the hosted OpenReception service this is a scope-changed escalation: a single customer-side tenant administrator gains full platform-wide administrative control over all other tenants' configuration, users, staff records, operational metadata, and tenant lifecycle. Plaintext appointment contents remain subject to the E2E model unless chained with the staff-crypto poisoning issue (V-4) or with staff-passkey hijacking (V-1). On a single-tenant self-hosted deployment it is still a privilege escalation because TENANT_ADMIN should not be able to create new tenants, modify global configuration, or manage other administrators. The same handler also accepts updates targeted at any colleague within the tenant. A tenant admin can promote a separate collaborator account instead of themselves, leaving their own audit trail clean while the platform-wide breach happens through a separate identity. Version 1.0.2 fixes the issue.

Subscriber SQL Injection in WP Job Portal <= 2.5.6 versions.

OpenReception's appointment booking software provides an end-to-end encrypted appointment booking platform. Prior to version 1.0.5, the unauthenticated `/api/tenants/{id}/schedule` endpoint returns every non-archived channel for a tenant regardless of the channel's `isPublic` flag. Channels marked `isPublic = false` are intended to be invisible to public callers; the dashboard creates them deliberately to hide internal-only services from the patient booking UI. The schedule endpoint ignores the flag entirely and discloses channel names, descriptions, IDs, agent associations, pause status, confirmation requirements, and computed slot availability for the requested date range. The asymmetry between `addAppointmentToTunnel` (which enforces `eq(channel.isPublic, true)`) and the schedule endpoint (which does not) confirms the design intent: private channels exist as a real access boundary in the booking flow, just not in the schedule disclosure. Version 1.0.5 patches the issue.

Unauthenticated PHP Object Injection in Adrena <= 1.2.14 versions.

An improper access control check in CatchPulse's named pipe communication interface could allow an attacker to invoke CatchPulse functions. This is limited to operations that enforce more restrictive security policies.