Cross-site request forgery in boneIO Black controllers lets a malicious webpage drive state-changing API calls - rebooting the device, toggling relay circuits, moving gates and blinds, or shutting it down - against any reachable unit. All versions before 1.6.0 are affected, and because authentication is disabled by default (CVE-2026-77053), any user on the same internal network who visits an attacker-controlled page can trigger those actions; where authentication is enabled, the missing CSRF tokens and Origin/Referer checks still allow forged requests to ride an existing session. Exploitation is unauthenticated (PR:N) but requires the victim to open the malicious page (UI:A), no public exploit has been identified at time of analysis, and the vendor fixed the issue in 1.6.0.
Out-of-bounds read in NXP's MCUXpresso SDK USB middleware (mcuxsdk-middleware-usb) before 2.13.0 allows a malicious USB host to disclose device memory by sending RNDIS SET messages for OID_GEN_CURRENT_PACKET_FILTER or OID_802_3_MULTICAST_LIST with an out-of-range informationBufferOffset. The CDC RNDIS device-class handler USB_DeviceCdcRndisSetCommand() validated informationBufferLength but trusted the host-supplied offset, so the read can run beyond the fixed command buffer. Exploitation requires physical USB access to an affected MCU that exposes the RNDIS function to an untrusted host; no public exploit identified at time of analysis, and the issue is not listed in CISA KEV.
Stored cross-site scripting in the Webasyst Framework Photos app (through 4.4.0) allows an authenticated backend user with album edit rights to save a crafted 'url' value that executes JavaScript in an administrator's backend session when the album settings dialog is opened. The issue is rated 5.1 (medium) and requires victim interaction; it is not remotely exploitable without a backend account and album-edit permission. No CISA KEV entry was identified and no public exploit identified at time of analysis, and the available data does not list a fixed version.
Server-side request forgery in Kameleo's proxy-validation feature affects versions up to and including 5.0.1.26 and lets an authenticated low-privilege user who interacts with the proxy tester manipulate the host and port arguments of InternalController.TestProxy so the client-engine application issues requests to destinations of the attacker's choosing. The assessed impact is limited (CVSS 5.1; low confidentiality, integrity, and availability impact), and exploitation requires access to the Kameleo client-engine API plus passive user interaction. No public exploit code has been identified at time of analysis and there is no CISA KEV entry; upgrading to 5.2.0 removes the vulnerable route.
NocoDB through 2026.08.0 has a medium-severity race condition in its password-reset flow, implemented in packages/nocodb/src/services/users/users.service.ts, with low impact on confidentiality, integrity, and availability. Exploitation is remote but requires a low-privileged authenticated account and interaction with the reset flow; the exact trigger sequence is not described in the available sources. No public exploit was identified at time of analysis, no fixed version is listed, and the vendor deleted the referenced GitHub issue without explanation.
A race condition was addressed with additional validation. This issue is fixed in macOS Golden Gate 27.2, macOS Sequoia 15.8.2, macOS Tahoe 26.7.2. An attacker may tamper with code-signed applications to execute arbitrary code in the application's context.
Out-of-bounds write in ExLlamaV2 0.3.0 through 0.3.2 can corrupt process memory when a crafted GPTQ-quantized model supplies manipulated g_idx index data that is processed by QMatrix::make_sequential in the CUDA extension. The issue is local-only and requires the victim to load an attacker-controlled model; impact is limited to integrity and availability with no confidentiality loss, and no public exploit was identified at time of analysis (EPSS 0.15%, low real-world priority).
A heap out-of-bounds read in libfvde's FileVault 2 encrypted metadata parser (metadata type 0x001a) can crash the application that opens an attacker-supplied volume, with impact limited to availability of the parsing process rather than data disclosure or code execution. The defect affects libfvde builds through the 20260923 snapshot and is reached when embedded XML plist data in an encrypted metadata block is not NUL-terminated, so libfvde_volume_open() executes an unbounded strlen that walks past the allocated heap block. The published CVSS 4.0 score is 4.8 (an assessed CVSS 3.1 vector of AV:L/AC:L/PR:N/UI:R/S:U/C:N/I:N/A:L matches the evidence), EPSS is 0.13% at the 3rd percentile, the issue is not listed in CISA KEV and is not confirmed actively exploited, and although a third-party technical write-up is referenced, no verified working exploit code is confirmed; realistic exposure is confined to forensic/DFIR tools and automated triage pipelines that ingest untrusted disk images.
Heap out-of-bounds read in gbsplay's VGM loader (vgm_open() in gbs.c) allows a crafted .vgm file to walk the command-stream parser past the end of its heap buffer and crash the player. All builds through 0.0.103 are affected; the trigger is a VGM stream containing wait opcodes without the 0x66 end-of-data terminator, which leaves the parse loop running to and beyond the buffer boundary. The vector is local and requires the victim to open or play the malicious file (CVSS:3.1/AV:L/AC:L/PR:N/UI:R), and impact is confined to availability - no code execution or confirmed disclosure. EPSS is 0.13% (3rd percentile), there is no KEV entry or confirmed active exploitation, no verified working exploit code is publicly available (a public technical write-up exists but is not verified as working exploit code), and a bounds-checking upstream fix commit is referenced.
Heap out-of-bounds read in libdwarf through 2.3.3's .debug_addr parsing path: dwarf_debug_addr_table() validates the unit length without including the table's starting offset, so an object file whose .debug_addr section declares a non-zero table offset slips past the size check and a later dwarf_debug_addr_by_index() call reads past the heap buffer. Only libdwarf-based consumers that actually parse an attacker-supplied object file are affected (local attack vector, user interaction required), and the outcome is bounded to disclosure of adjacent heap bytes or a crash. No confirmed active exploitation (not in CISA KEV) and EPSS is only 0.13% (3rd percentile); a public write-up and PoC reference exist but are not verified as working exploit code, and an upstream fix commit is available.
Heap out-of-bounds read in gbsplay 0.0.103 and earlier lets anyone who opens a crafted GBR file sized between 0x21 and 0x162 bytes read past the end of a heap buffer while gbr_open() in gbs.c scans the ROM title field, crashing the player or leaking a small amount of adjacent heap memory. The flaw is reachable in the default configuration with no special option enabled, but it requires the victim to load the attacker-supplied file locally, so it is a low-priority local issue with no code execution or integrity impact (CVSS 4.0 base 4.8; independent assessment CVSS 3.1 AV:L/AC:L/PR:N/UI:R, C:L/A:L; EPSS 0.13%, 2nd percentile). Vulnerability is not in CISA KEV and SSVC records exploitation as none; a public technical writeup describing the flaw exists, and an upstream fix commit is available.
Heap out-of-bounds write and read in HTSlib through 1.24 is triggered when a crafted .gzi BGZF index stores an entry count of 0xFFFFFFFF, which truncates the 32-bit noffs/moffs fields in bgzf_index_load_hfile() to zero and collapses the index allocation to malloc(0); subsequent writes past that buffer and an offs[-1] read in bgzf_useek() corrupt heap memory or crash HTSlib-based tools such as samtools and bcftools. The vector is local: the victim must load an attacker-supplied .gzi index file and then perform a seek, with assessed impact limited to Low integrity and Low availability - no confidentiality impact and no confirmed code execution. EPSS is 0.12% (2nd percentile), there is no CISA KEV listing, and a public technical write-up describing the truncation exists; it is not verified as working exploit code.
Kameleo through 5.0.1.26 can disclose profile information to a local, low-privileged user through the Profile Controller's ListProfiles function (CWE-200). The supplied CVSS 4.0 vector scores 4.8 (medium) with confidentiality-only impact; the vendor fixed it in 5.2.0, and the supplied intelligence does not confirm CISA KEV exploitation or working public exploit code.
vLLM's OpenAI-compatible server before 0.32.0 registers Starlette CORSMiddleware with allowed_origins defaulting to ["*"], so a malicious web page loaded by a victim can issue cross-origin requests to a vLLM server the browser can reach and read the responses. When no --api-key or VLLM_API_KEY is configured - the default - endpoints such as GET /v1/models and POST /v1/chat/completions are reachable, letting the page enumerate models, drive chat completions, burn GPU time, and read model output through the victim's browser; if a key is set, the guarded prefixes require that the caller already hold a usable token. Cookie-based CSRF does not apply because allow_credentials is False by default, and exploitation needs user interaction plus a browser-reachable server, which is why this scores only 3.4 and is tracked as monitor priority; no public exploit code was identified at time of analysis.
Script injection in seroval's JavaScript-output serializers lets a RegExp whose source contains '<' terminate an HTML <script> element and execute attacker script in the rendered page, affecting releases before 1.0.0 and from 1.5.5 up to but not including 1.6.10. The flaw only materializes in applications that build regular-expression syntax from untrusted input, serialize that RegExp through serialize, serializeAsync, crossSerialize, crossSerializeAsync, crossSerializeStream, Serializer or compileJSON, and embed the output inside a script element as server-side rendering commonly does; a victim must then load that page (CVSS:3.1/AV:N/AC:H/PR:N/UI:R/S:C/C:L/I:L/A:N, 4.7). EPSS is low (0.19%, 8th percentile), the issue is not in CISA KEV, and no public exploit has been identified at time of analysis, though the vendor has released 1.6.10 with the fix.
BRAIN2 installations up to and including version 3.09 can write the cleartext password of the Windows service account that runs the BRAIN2 MainService into a setup log file, but only when Microsoft SQL Server access is configured with Integrated Security and the service account is specified during the installation wizard. The log lands in the temporary directory of the administrator who ran the installer, which the vendor states is normally restricted to administrative users, so an attacker needs local access and effective administrative rights to reach the credential. CVSS 3.1 scores this 4.4 (AV:L/AC:L/PR:H/UI:N/S:U/C:H/I:N/A:N); no public exploit code has been identified at time of analysis and the issue is not in CISA KEV.
Image handling on unpatched Apple macOS releases can be driven into a buffer overflow that terminates the processing process, giving a denial-of-service outcome with no confidentiality or integrity impact. The flaw is addressed by improved bounds checking in macOS Golden Gate 27.2, macOS Sequoia 15.8.2 and macOS Tahoe 26.7.2, so only systems below those builds are exposed; exploitation requires the victim to process an attacker-supplied maliciously crafted image (CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:U/C:N/I:N/A:L, 4.3). No public exploit identified at time of analysis, SSVC records no known exploitation and EPSS is 0.48% (41st percentile), making this a this-month, low-urgency patch item rather than an emergency.
Missing authorization in the VillaTheme Customer Coupons for WooCommerce plugin, affecting versions through 1.3.14, allows an authenticated low-privilege user such as a registered customer to perform a sensitive coupon-related action without the required permission check. The flaw is rated CVSS 4.3 (Medium) with limited integrity impact only-no confidentiality or availability impact-and no public exploit has been identified at time of analysis. The trigger endpoint and exact action are not described in the available sources; exploitation requires a WordPress site running WooCommerce and the vulnerable plugin plus a valid low-privilege account.
Zephyr's Bluetooth host (subsys/bluetooth/host, versions 2.3.0 through 4.4.2) can be driven to permanently strand connection reservations: when CONFIG_BT_EXT_ADV_MAX_ADV_SET > 1 and an application runs a directed and an undirected connectable extended advertising set on the same identity, an adjacent, unauthenticated attacker who freely chooses an initiator address mapping to the directed set's target causes the wrong bt_conn to be consumed, leaving the undirected set's reservation unreleasable. Each occurrence burns one CONFIG_BT_MAX_CONN slot until bt_disable() or reboot, and repeated attempts against applications that restart the directed set after each disconnect exhaust the pool so connectable advertising fails with -ENOMEM; the directed set also keeps advertising without a reservation, so genuine connections on it are dropped. Impact is availability only (no memory corruption or information disclosure), the vulnerable configuration is non-default on both Kconfig options, and no public exploit code was identified at time of analysis and the issue is not in CISA KEV.
Missing authorization in APIExperts Square for WooCommerce through 4.7.5 lets an authenticated low-privilege user (PR:L) perform limited unauthorized changes, with CVSS 4.3 and integrity-only impact. The flaw is in the WordPress plugin slug woosquare and is classified CWE-862; no public exploit code is identified at time of analysis. No fixed version is specified in the available advisory data, so sites should monitor the vendor/Patchstack record for an update.
Cross-site request forgery in the WordPress plugin Media Library Assistant through version 3.42 lets an attacker induce a logged-in site user to submit a forged request in that user's existing session, producing low-integrity changes with no confidentiality or availability impact. This is a moderate-severity, low-priority issue rather than an urgent one: the attacker needs no account or privileges of their own but cannot succeed without a victim authenticated to WordPress who is tricked into visiting attacker-controlled content. No public exploit code or confirmed active exploitation (CISA KEV) was identified at time of analysis.
Missing authorization in the VillaTheme DEPART WordPress plugin (depart-deposit-and-part-payment-for-woo) through version 1.0.10 allows an authenticated low-privilege user to perform an action that should require higher permissions, resulting in limited integrity impact on the affected WooCommerce store. The flaw is rated CVSS 4.3 (CWE-862) with network reachability and no user interaction required, and it only affects sites where the plugin is installed and active. No public exploit identified at time of analysis, and the risk signals (EPSS/KEV absent, patch-level priority 'this_month') point to a moderate, non-urgent issue rather than an emergency.
Missing authorization in the Meow Apps AI Engine plugin for WordPress (versions from n/a through 3.8.6) lets an authenticated low-privilege user - for example a subscriber on a site that allows self-registration - perform actions that should require higher capabilities, with impact limited to integrity modification. The flaw is remotely reachable and low-complexity (CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:L/A:N, 4.3) but requires an existing account, and there is no confidentiality or availability effect. No public exploit code has been identified at time of analysis; the issue is tracked for remediation this month rather than as an emergency.
Private annexed content can be exfiltrated from a victim's git-annex repository through attacker-controlled remotes, allowing data the user believes is private (annexed files kept out of a published git repo, or objects in a private S3 bucket or other special remote) to be pulled and sent to the attacker. The assessed vector is CVSS:3.1/AV:N/AC:H/PR:N/UI:R/S:C/C:H/I:N/A:N: remote, unauthenticated in the sense that no credentials to the victim's systems are needed, but requiring user interaction and three preconditions from the vendor advisory - the attacker must control one of the victim's remotes, must already possess reconnaissance about the private repository (specific annexed keys, or the location of the private git repository), and the victim must run a content-transferring command such as `git-annex sync --content` against a special remote autoenabled on init or manually enabled and then retargeted. Versions before 10.20261026 are affected; no CISA KEV listing, no public exploit code, and no EPSS score were available at time of analysis, and the vuln.today priority label is 'awaiting_data', so exploitation has not been observed.