Case-sensitivity defects in MISP Guard's feed-distribution policy checks allow an authenticated MISP event creator to slip events past configured TLP taxonomy, allowed-tag, blocked-attribute-category and blocked sharing-group rules simply by changing the letter case of a tag, category or UUID string, with the result that threat-intelligence data the operator meant to restrict is still forwarded to feed consumers. Only deployments running misp-guard 1.3 or earlier that actually rely on these allow/block policies are exposed, and the impact is confidentiality-only - no integrity or availability effect. No public exploit code was identified at time of analysis; the vendor has published an upstream fix commit (9868b1d9) that normalizes case on both sides of every affected comparison and adds regression tests reproducing the bypass.
Local-network attackers can cause unexpected system termination or process memory corruption on Apple macOS versions before 15.8.2, 26.7.2, and 27.2 by triggering a double-free memory-management bug. The issue is fixed in macOS Sequoia 15.8.2, macOS Tahoe 26.7.2, and macOS Golden Gate 27.2, and no public exploit has been identified at time of analysis; EPSS is low at 0.35% and CISA KEV status is not indicated.
A NULL pointer dereference in libcss through 0.9.2 lets remote attackers crash any application that parses an attacker-supplied stylesheet whose @media or @import media query list ends in a trailing comma, taking down NetSurf and other embedders of the library. The fault sits in mq_parse_media_query() (src/parse/mq.c), where the final iteration of the media-list loop hands a NULL token to a branch that dereferences it unchecked; the regression tests added with the fix ('@media),', '@media ),,,', '@media (color),') confirm the exact trigger. Impact is availability only - a client-side parse-time crash with no code execution or data disclosure, and it requires the victim to load the hostile CSS (passive user interaction). No confirmed public exploit code was identified at time of analysis, though a public technical writeup describing the dereference exists; EPSS is low at 0.33% (26th percentile) and there is no CISA KEV listing, so this is a moderate-priority denial-of-service issue rather than a top-tier emergency.
YesWiki releases up to and including 4.6.7 let a self-registered, low-privileged account read wiki pages and Bazar entries whose read ACLs are restricted to other users or groups. The flaw is a read-ACL bypass: AclService builds its access filter with a SQL LIKE pattern built from the account name, and the escaping applied (mysqli_real_escape_string) neutralises quotes but not the LIKE metacharacters _ and %, so a name such as '___' or '%%%' over-matches ACL tokens it was never granted. Exploitation is confirmed to require only self-registration (on by default) plus a call to the public Bazar listing API; the upstream advisory includes reproduction steps but no independently verified working exploit code is referenced, EPSS is low at 0.28% (20th percentile), and there is no CISA KEV entry. Impact is confidentiality-only - no write or availability effect was demonstrated, and the practical exposure depends on access-restricted Bazar entries actually existing on the target.
OpenPanel before 3.0.0 leaks secrets into standard-output logs through two separate code paths, allowing anyone who can read application, container, CI or centralized logs to recover sensitive data. Database migration startup banners print the full DATABASE_URL and CLICKHOUSE_URL including user-info credentials and credential-like query parameters, and the email fallback path used when both SMTP_HOST and RESEND_API_KEY are unset serializes the recipient address, subject and full template data - which for the password-reset flow includes the reset-token URL. Exploitation is not remote against the service itself: it requires pre-existing read access to log stores plus, in the migration case, that the code-migration command was run, or, in the email case, the non-default double-unset provider configuration and a triggered email flow; no public exploit code is identified at time of analysis and EPSS is low (0.28%, 20th percentile), consistent with a moderate-priority information exposure rather than an emergency.
SQL injection in OpenPanel's conversion analytics query allows project-read members and share-link holders to read other tenants' ClickHouse analytics data. Before version 3.0.0, ConversionService.getConversion interpolates series event names and startDate/endDate values raw inside single quotes, so a payload such as x' OR 1=1 neutralizes the project_id tenant scoping; the projectId value itself is passed through sqlstring.escape, so the omission is confined to the event-name and date fields. Exploitation requires an authenticated low-privilege actor (CVSS 4.0 PR:L) or a valid share link, and cross-tenant impact depends on a multi-tenant deployment in which the ClickHouse reader credential can read all rows. The advisory includes a PoC payload description that is not verified as working exploit code; SSVC records exploitation as none and EPSS is 0.23% (14th percentile), so this is a real but moderate-priority exposure focused purely on confidentiality.
SQL injection in OpenPanel's cohort preview query lets any user holding project-read access on any project read ClickHouse data belonging to other tenants; this is a genuine but access-gated data-disclosure issue rather than a drop-everything emergency. The flaw is confined to the ClickHouse analytics backend (the default) in multi-tenant deployments, requires a valid low-privileged account (authenticated, project READ only, no write or admin role), and affects OpenPanel versions before 3.0.0. Exploitation permits read-only cross-tenant disclosure with no integrity or availability impact, there is no CISA KEV listing, EPSS is low at 0.23% (14th percentile), and a public PoC reference exists in the vendor advisory (a cohort.preview payload using operator isNotNull with an empty value array); it is not verified as working exploit code. A vendor patch is available, and the source-level fix is to apply the PROFILE_COLUMNS allowlist to profileColumnAccess.
Unauthenticated screen-content disclosure in WolfVision Cynap Pro and Cynap Core Pro devices with vSolution Matrix enabled lets hosts on the same network segment view or capture the current presentation through an unencrypted RTSP/RTP stream. The issue affects versions below 1.78a and is limited to confidentiality; the vendor has released 1.78a, and no public exploit was identified at time of analysis. Exposure is bounded by the non-default Matrix feature requirement and network adjacency.
Bluetooth Object Transfer Service (OTS) servers in Zephyr (>= 3.3.0 through <= 4.4.2) mishandle failed Calculate Checksum responses: oacp_ind_send() appends a four-byte Checksum Value field based only on the requested op code, then calls net_buf_simple_pull_le32() on a zero-length stack buffer whenever validation failed. Any peer within radio range can abuse this with a nine-byte OACP write and no pairing, encryption or authentication, because the OACP characteristic and its CCC are exposed with unauthenticated write permissions. On CONFIG_ASSERT builds the zero-length pull trips the assertion in net_buf_simple_pull_mem() and panics the device (remote DoS); with assertions compiled out the same path copies four bytes of uninitialized stack memory into the indication sent to the peer, repeatable per request and bounded to four bytes with no memory corruption; the reporter's reproducer demonstrates the panic and no public exploit tooling beyond it is confirmed.
The Mosquitto MQTT broker baked into boneIO black_debian_images uses one shared default username/password across every device, and those credentials are published on GitHub with no instruction to change them. Any host that can reach the broker on the same LAN/Wi-Fi segment can connect unauthenticated in the CVSS sense (PR:N - no prior account or privilege needed) by simply presenting the published defaults, gaining full read and write access to the broker and therefore control over connected relays, blinds and gates, plus the ability to inject fake entities into a linked Home Assistant instance. Version 1.6.0 generates per-device MQTT passwords at first boot, but devices upgraded from earlier images keep their old broker credentials, so the flaw persists until the owner manually changes them; no public exploit code was identified at time of analysis, and the CERT-PL report is prioritised as 'this_week'. The vendor's CVSS 4.0 score is 7.1 (AV:A/AC:L/PR:N/UI:N/VC:H/VI:N/VA:N), which understates the described write/integrity and potential physical-safety consequences.
Cross-site request forgery in the Webasyst Framework Photos app through 4.4.0 allows a malicious page to submit forged POST requests as a logged-in backend user, deleting photos and changing access rights. The Photos app does not enable the framework's opt-in CSRF check, and exploitation requires the victim to hold an active authenticated session and to be induced into loading the attacker's page; the attacker needs no privileges of their own (PR:N), but the victim interaction prevents unattended mass exploitation. Impact is confined to the photo/delete, album/save, photo/manageAccessSave and photo/saveField endpoints with high integrity impact and no confidentiality or system-level effect; this is a moderate, not emergency, priority and no public exploit was identified at time of analysis.
Blind SQL injection in the Mautic dashboard widget data API (/api/data/{type}) allows an authenticated account holding API access to read arbitrary contents of the Mautic database, including contact and marketing records and user accounts with password hashes. Exploitation requires two conditions to hold simultaneously: the Mautic API must be enabled on the instance, and the attacker must hold a valid account with API access (CVSS PR:L); no unauthenticated path exists, and instances with the API disabled or restricted to trusted users are not exploitable. Integrity impact is none - data modification is not possible - while availability impact is limited to attacker-induced query slowdowns; no public exploit code or CISA KEV entry was identified at time of analysis, and the vendor has released fixed versions 7.2.2, 6.0.10 and 5.2.12.
Authenticated SQL injection in Mautic's contact lookup AJAX endpoint (/s/ajax) allows any valid user account, regardless of role or assigned permissions, to run blind boolean-based or time-based queries against the Mautic database and read data the database user can access, including user email addresses and password hashes, contact records, and campaign and email content. The flaw affects Mautic releases from 2.1.0 before 5.2.12, 6.0.x before 6.0.10, and 7.0.x before 7.2.2; anonymous access is not sufficient because a valid account is required. No public exploit code has been identified at time of analysis, and the priority signal is this_week rather than an emergency.
Missing authorization in Frappe CRM before 1.89.1 lets an authenticated low-privileged user read CRM Lead records they are not permitted to see by supplying a known phone number to the crm.integrations.api.get_contact_by_phone_number endpoint, which resolves matches with raw database queries instead of the permission system. Any user holding a valid session and a low-privilege role (the patch's tests use a 'Sales User' account) can enumerate phone numbers to disclose lead details otherwise blocked by standard API and UI controls; no public exploit was identified at time of analysis and the issue is not in CISA KEV. Severity is high for confidentiality (CVSS 7.1 in the 4.0 scoring, with the assessed 3.1 vector at AV:N/AC:L/PR:L/UI:N/C:H) but the impact is disclosure-only and requires a known or guessed phone number per record, which keeps this a moderate-priority, patch-this-week item rather than an emergency. The vendor shipped a fix in v1.89.1.
Out-of-bounds reads in the SOEM EtherCAT master library (2.0.0 and earlier) let a malicious or compromised device on the same EtherCAT segment make the master read past the end of allocated buffers, disclosing memory contents or crashing the master process. No authentication or user interaction is required, but the attack surface is confined to the adjacent Layer-2 fieldbus segment - typically an isolated OT control network rather than anything internet-routable. No public exploit code or confirmed in-the-wild exploitation is identified in the available data, and the vuln.today decision label is this_week, reflecting a genuine but bounded industrial/OT priority rather than an internet-wide emergency.
OpenPanel deployments before 3.0.0 are exposed to SQL injection through the tRPC `event.events` query, which splices an unconstrained `profile.*` filter name directly into the profiles-JOIN subquery, letting any authenticated user with read access to a single project comment out the `WHERE project_id` scoping and read events (path, country, city, profile_id, revenue) from every other tenant. The same unsanitized JOIN construction also appears in `getEventsCount` and the funnel service, so the exposure is not limited to a single endpoint. A public PoC reference with reproduction steps accompanies the GitHub advisory (GHSA-pc3q-gw7f-p2x2); it is not verified as working exploit code and there is no KEV entry. EPSS is low at 0.27% (19th percentile), and the vendor fixed the flaw with an allowlist in the 3.0.0 release.
SQL injection in the @aws-cdk/aws-redshift-alpha database-query custom resource handlers, affecting versions before 2.274.0-alpha.0, lets an actor who can influence construct property values - specifically the user name, comment, or password fields - execute arbitrary SQL against Amazon Redshift with superuser privileges, including granting access to all cluster users. The assessed vector is local with low privileges and required user interaction: a CDK stack deployment must actually run with the attacker-influenced values, so this is not remotely reachable against a running service and deployments built only from trusted, static property values are not realistically exposed. AWS has released 2.274.0-alpha.0 to address the flaw, and no public exploit code has been identified at time of analysis.
Reachable assertion in llama.cpp (through 0.7.0) lets unauthenticated remote callers abort the inference process by driving the non-default Mirostat sampler into non-finite or zero-sum token probabilities, producing a denial of service that is confined to availability. Only deployments that expose an inference endpoint such as llama-server to untrusted clients - and that permit those clients to select Mirostat v1/v2 and influence sampling parameters or inputs that make probabilities non-finite - are affected; there is no code execution, data disclosure or integrity impact. No public exploit code and no CISA KEV entry were identified at time of analysis, and the issue carries a moderate this_month priority rather than urgent handling. The flaw was reported upstream through pull request #29992, which adds the missing guards; a tagged release containing that fix is not independently confirmed.
Unauthenticated PHP file inclusion in WWBN AVideo (confirmed affected: release tags v29.3.0 and v29.4.0 plus master snapshot 8a32cc70a4758c74c54cdc80c592f11514debcb2) lets a crafted Accept-Language header steer the locale loader outside the locale/ directory by smuggling overlapping traversal ('....//') past a single-pass str_replace('../','') filter, causing include_once to execute PHP files that already exist elsewhere in the web root. The inclusion primitive needs no account, privilege or victim interaction when browser-based automatic language detection or browser-language fallback is active, but the resulting impact is bounded by which readable PHP files already exist at a reachable path - arbitrary attacker code execution would require a separate, undemonstrated ability to upload a PHP file, and the advisory did not show secret disclosure, arbitrary file reading or quantified availability loss. No public exploit code was identified at time of analysis, EPSS is 0.51% (43rd percentile) and the issue is not in CISA KEV, so the decision label is this_month rather than act_now.
Heap-based buffer overflow in NXP MCUXpresso SDK USB middleware before 2.13.0 lets a malicious USB device exploit unvalidated bNumEndpoints and bLength descriptor fields in the host audio, video, and CDC class drivers, writing past the fixed-size endpoint list and corrupting heap memory. The affected component must run as a USB host with one of those class drivers active, and the attack requires physical attachment of the crafted device (CVSS 4.0 AV:P); no remote path is described. Successful exploitation can lead to code execution or denial of service, but no public exploit was identified at time of analysis and KEV status is not indicated.
Unauthenticated code injection in DedeCMS 5.7.118 lets a remote attacker manipulate the 'aid' argument of the advertisement-rendering endpoint plus/ad_js.php, which is processed through the _RunMagicQuotes function in common.inc.php, to inject code into the application (CWE-94). The CVSS 4.0 and independent assessments agree on network reachability with no privileges and no user interaction (AV:N/AC:L/AT:N/PR:N/UI:N), while impact sub-scores are Low, indicating a partial rather than full compromise is the likely ceiling. No public exploit code has been identified at time of analysis, the issue is not in CISA KEV, and the vuln.today decision label places it in this_month - routine triage rather than emergency response, with exposure limited to installations where plus/ad_js.php is reachable over the network.
Information exposure in the a2a-go SDK for the A2A protocol (versions up to and including 2.4.0) causes push notification configuration data to be written into an error message returned by factory.CreateExecutor in a2asrv/agentexec.go. The leak only occurs when saving the push configuration fails, at which point the pre-patch code interpolated the entire params.Config.PushConfig structure into the returned error, exposing callback/webhook details to the remote caller; the assessed impact is confidentiality-only (C:L) with no integrity or availability effect, reachable remotely with no authentication or user interaction required per the CVSS vector. No public exploit code or PoC is identified in the supplied sources and the issue is not in CISA KEV; the vendor fix is a2a-go v2.5.0 (PR #393, commit f4d5882).
An out-of-bounds read in nDPI's SoftEther dissector (dissect_softether_host_fqdn() in softether.c, all versions through 6.0) lets remote, unauthenticated attackers crash or destabilize a DPI engine that is inspecting traffic. The flaw is only reachable once nDPI has already classified a flow as SoftEther: the attacker then sends a crafted short or truncated SoftEther UDP datagram so that the consumed byte count passed to strncmp exceeds the real payload length, and the comparison runs past the end of the packet buffer. Impact is limited to availability of the monitoring process (no confidentiality or integrity effect), and the precondition of an existing classified SoftEther flow constrains opportunistic abuse; EPSS is low at 0.43% (37th percentile), there is no KEV listing, and no working exploit code appears in the available data, although a public technical write-up of the bug exists.
Connection-capacity exhaustion in the MICE (Miracast over Infrastructure Connection Establishment) service on WolfVision Cynap collaboration devices lets an unauthenticated remote attacker with network reachability to that service occupy its available connection slots, so legitimate users cannot establish MICE-based Miracast sessions while the condition persists. Exposure is bounded: MICE is disabled in the default configuration and must be explicitly enabled by the customer, and Miracast P2P is unaffected and remains available independently of MICE on most devices, although a complete Miracast disruption can occur on some units when WLAN is configured for Infrastructure mode. EPSS is 0.31% (23rd percentile), no public exploit has been identified at time of analysis, and vendor patches are available for the version branches that list fixed releases.
Parsing deeply nested JSON through go-jose versions prior to 4.1.5 can exhaust Go's stack limit and crash the process, giving remote attackers a denial-of-service primitive against any service that feeds untrusted JWT, JWE, JWS, or raw JSON into the library. The CVSS 4.0 vector rates it AV:N/AC:L/PR:N/UI:N with availability-only impact (score 6.9), while our independent CVSS 3.1 assessment scores it 6.7-level severity as A:L - the practical exposure is limited to applications that actually expose a go-jose parsing path to external input. EPSS is low at 0.29% (22nd percentile) and no public exploit has been identified at time of analysis; the vulnerable v4 branch is fixed in 4.1.5, but the v3 branch (<= 3.0.5) remains unpatched.
Unauthenticated access to the Bull Board queue dashboard in OpenPanel self-hosted deployments before 3.0.0 exposes queue administration on the worker HTTP port: anyone who can reach that port can load the dashboard, list every BullMQ queue (with readOnlyMode:false and allowRetries:true), read queue data, Prometheus metrics and Redis/PostgreSQL/ClickHouse health, and call mutating Bull Board endpoints such as add, retry and clean. Exposure is deployment-dependent - the default docker-compose.template.yml does not publish the worker port, but the shipped Coolify template assigns SERVICE_FQDN_OPBULLBOARD to the worker and thereby publishes the UI on a public hostname, and the reporter's test instance bound the port on all interfaces. The vendor advisory includes PoC requests demonstrating HTTP 200 on GET / and GET /api/queues with no 401, but a crafted privileged job add was not demonstrated; EPSS is 0.27% (19th percentile) and there is no CISA KEV entry, and the vendor fixed the issue in 3.0.0.
Decrypting untrusted JWE ciphertext in Go JOSE (go-jose) with CBC-HMAC content encryption can pass an empty buffer into CBC-HMAC unpadding, causing a panic and denial of service. Applications using ParseEncrypted, ParseEncryptedJSON, or ParseEncryptedCompact with A128CBC-HS256, A192CBC-HS384, or A256CBC-HS512 on attacker-controlled ciphertext are affected; the v4 line is fixed in 4.1.5, while the advisory lists v3 through 3.0.5 as not yet patched. No public exploit was identified at time of analysis; EPSS is 0.26% (18th percentile), and vuln.today labels the priority this_month.
Unauthenticated information exposure in WOLFVISON Cynap collaboration and presentation devices allows anyone who can reach the web interface to retrieve resources that should only be reachable locally on the device, including screen-saver content and web-interface assets. The flaw stems from a broken access-control check for remote requests and affects Cynap Pro, Cynap Core Pro, Cynap Receiver, Cynap Pure, Cynap Pure Mini, Cynap Pure Pro before 1.78a, Cynap Videobar before 1.72e, and Cynap Pure (Next Gen) and Cynap Bridge before 2.00d, with the Cynap/Cynap Core/Cynap Pure SDM line listed as affected without a fixed version in the available sources. Per the vendor the exposed resources carry no credentials and no administrative capability, EPSS is low at 0.26% (17th percentile), and no public exploit or active exploitation is indicated, so priority is rated this_month rather than urgent.
Blind SQL injection in the YesWiki despam action (tools/security/actions/despam.php) allows an authenticated administrator to read arbitrary data from the wiki database, including rows from yeswiki_users such as password hashes, affecting every release before 4.7.0. The 'from' POST parameter is passed through mysqli_real_escape_string() but then lands in an unquoted INTERVAL operand, so escaping is inapplicable and attacker-supplied parentheses, keywords and comment markers are interpreted as SQL. Exploitation requires a YesWiki administrator account (despam.php is gated on UserIsAdmin() at line 23); a public proof-of-concept script is published in the vendor's GitHub security advisory and was verified against the doryphore-dev branch, with the result-row count and SLEEP() timing both serving as blind oracles. EPSS is low at 0.25% (16th percentile) and there is no KEV entry; the vuln.today decision label is 'this_month', reflecting a genuine but privilege-gated confidentiality issue rather than an emergency.
Unauthenticated cross-tenant disclosure in OpenPanel before 3.0.0 allows any remote attacker to read other tenants' project annotations - release notes, deploy markers and internal event notes - by calling the public reference.getChartReferences tRPC procedure with a guessed slug-based projectId. The handler never reads the session and performs no project-access or share-token check, so it returns title, description and date rows for any project whose ID the attacker can guess; impact is confidentiality-only (CVSS 6.9, assessed as CVSS:3.1 C:L with no integrity or availability effect). EPSS is 0.25% (16th percentile) and there is no CISA KEV listing, but the vendor advisory documents a public proof-of-concept request and the issue is fixed in 3.0.0.
Unauthenticated review forgery in QloApps 1.6.0 through 1.7.0 allows remote attackers to post fake ratings and reviews attributed to other customers by reusing the publicly exposed review_ajax_token and enumerating sequential id_order values against the qlohotelreview module's AddReview AJAX action. The impact is limited to integrity - forged review content and ratings attached to arbitrary orders - with no read of order data and no effect on availability, consistent with an assessed CVSS:3.1 vector of AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:L/A:N. No public exploit has been identified at time of analysis and EPSS is low at 0.25% (16th percentile), although the attack is automatable and SSVC records no observed exploitation; the vendor has published an upstream fix.
Memory-exhaustion denial of service in the erlang-jose JOSE/JWE library for Erlang and Elixir (versions 1.0.0 through 1.11.12) lets a remote, unauthenticated attacker inflate a few-kilobyte JWE token into hundreds of megabytes or gigabytes on the decrypting node and crash the BEAM. When a token carries the optional zip: "DEF" header, jose_jwe_zip:uncompress/2 feeds the entire attacker-controlled plaintext to zlib:inflate/2 with no output-size or inflation-ratio bound, and this happens automatically inside jose_jwe:block_decrypt/2 (JOSE.JWE.block_decrypt/2) before the recipient can inspect or reject the payload; with RSA-OAEP or ECDH-ES key management the attacker only needs the recipient's public key and no secret material. Exposure is limited to applications that actually decrypt JWEs received from untrusted senders and do not reject compressed headers first; EPSS is 0.25% (16th percentile), no confirmed active exploitation (CISA KEV) is recorded, and a proof-of-concept script with supporting logs is published in the upstream advisory GHSA-wfpv-qxp8-x5gm.
Pleasanter's LDAP authentication path can grant access when a request supplies a valid Login ID but no password, letting an attacker who knows an existing account identifier operate with that account's privileges. The issue affects Pleasanter deployments configured to use LDAP authentication; the CVSS 4.0 base score is 6.9 (CWE-304) with low confidentiality and integrity impact and no availability impact. No public exploit was identified at time of analysis, and the vendor and JVN advisories should be consulted for fixed-release and configuration details.
CPU-exhaustion denial of service in the erlang-jose library (versions 1.7.7 through 1.11.12) lets remote attackers who can get an application to parse an untrusted JWK pin BEAM schedulers by submitting a private RSA key that carries only d, e and n and omits every CRT parameter. Parsing such a key drives an internal prime-recovery loop that performs a modular exponentiation over the attacker-supplied modulus on every iteration; when the modulus is itself prime the search can never succeed and burns work until a one-second watchdog kills the worker, so roughly 211 bytes of input costs about one CPU-second. The advisory ships a public proof-of-concept script that was verified end-to-end to reach that watchdog, no credentials or valid RSA key are required, and sustained denial of service needs a flood of concurrent requests because the watchdog bounds each individual parse; EPSS is low (0.24%, 14th percentile) and there is no CISA KEV entry, consistent with a moderate, availability-only issue rather than a top-tier emergency.
Unauthenticated information disclosure in YesWiki 4.6.0 through 4.6.7 (and the 5.0.0 alpha pre-releases) lets anyone append the showreq parameter to a public Bazar listing request and receive the server-assembled SQL query, exposing table names with the configured prefix, JSON storage paths, ACL filter logic and database collation. The affected Bazar API routes are public by default, so default installations exposing an entry-listing endpoint or a {{bazarliste}} page are exploitable with no authentication, role, debug flag or user interaction; the echoed SQL is returned in the JSON body under a rawOutput key. This is reconnaissance-grade confidentiality loss only (no data, integrity or availability impact demonstrated), a public PoC walkthrough exists with source-level verification, EPSS is low at 0.23% (14th percentile), there is no KEV listing and no confirmed active exploitation, and the vuln.today decision is this_month with the vendor fix published in 4.7.0.
QloApps through 1.7.0 exposes guest order data to anyone who can guess a cart identifier: OrderConfirmationController::init() builds its guest-tracking redirect from the attacker-supplied id_cart without confirming that the requester holds the order's secure_key, so sequential id_cart values can be enumerated to harvest guest order references, email addresses and booking details, and to convert a guest into an account with an attacker-chosen password. Only guest-created carts are affected - registered-customer orders follow the history flow and are not exposed - and the impact is bounded to low confidentiality and low integrity loss rather than code execution or data destruction. Remote unauthenticated exploitation of default installations is possible with no user interaction; EPSS is low (0.23%, 13th percentile), no public exploit code has been identified at time of analysis, and the weakness is not listed in CISA KEV, so this fits scheduled remediation rather than emergency response.
Unauthenticated template-selection bypass in YesWiki before 4.7.0 lets remote attackers append a crafted `template` GET parameter to any page that uses the `{{bazarliste}}` action, forcing Bazar entries to render through an arbitrary template inside the `@bazar` Twig namespace. The parameter reaches Twig after only `htmlspecialchars()`, with no `basename()` or `hasTemplate()` check, so an attacker can swap layouts for every visitor who follows the crafted link, producing page defacement and user confusion. Impact stays within display manipulation: Twig's `validateName()` blocks namespace escape and filesystem reads, the effect is non-persistent, and no data modification or confidentiality loss was demonstrated. Risk signals are modest - EPSS 0.21% (11th percentile), SSVC exploitation 'none', not in CISA KEV, and a public proof-of-concept demonstration accompanies the advisory; no active exploitation is confirmed.
Unauthenticated bot-event disclosure in OpenPanel (all versions before 3.0.0) lets anyone POST to the event.bots tRPC procedure with only a projectId and receive bot-event rows - name, type, path and created_at - from projects whose overview share is private or password-protected. The procedure's anonymous branch only checks that a shareOverview row exists and never reads the share's public flag or password, so a share that was created and later toggled off (the share router has no delete path, so the row persists) still leaks, and the exposed path field can reveal internal routes. Impact is confidentiality-only against projects that have such a share row; EPSS is low at 0.21% (11th percentile), SSVC records exploitation as none, a public PoC request is documented in the advisory but is not verified as working exploit code, and the vendor has released a fix in 3.0.0.
Cleartext transmission of the Live UDP H.264 video stream in Yoosee GM8135S camera firmware 21.00.01.36 lets anyone positioned on the network path observe the endpoint at /npc/npc and read the unencrypted video feed, with confidentiality impact only (no integrity or availability effect). The camera requires no authentication (PR:N) and no user interaction (UI:N), but exploitation is rated high-complexity because it depends on first obtaining an on-path or shared-network vantage point (same LAN/Wi-Fi segment, upstream router, or other interception position) rather than simply sending a crafted request. A public PoC reference exists; it is not verified as working exploit code, no CISA KEV entry is associated with this CVE, and the vendor was contacted early but did not respond, so no patched firmware version is confirmed.
Out-of-bounds heap write in Furnace's FamiTracker FTM importer can crash the application and may allow limited memory corruption when a user opens a crafted module with two or more subsongs. Furnace through 0.6.8.3 is affected; exploitation is local-file only and requires user interaction, with the trigger in the GROOVES-block handling in DivEngine::loadFTM(). This is a low-priority issue under the supplied decision signals, with EPSS 0.14%, no CISA KEV entry, and no public exploit identified at time of analysis.
Heap out-of-bounds read in libphdi (all versions through the 20260902 snapshot) lets a crafted Parallels disk image descriptor crash any application that parses it, or expose a limited amount of adjacent heap memory. The bug sits in libphdi_data_files_get_extent_data_file_path and its wide-character twin: the extent data file name length is computed from the directory prefix length rather than the remaining basename length, so a long directory prefix paired with a short basename yields an over-long size and later reads walk past the heap allocation. The vector is local with a passive user-interaction step (the victim must open or process the malicious image), unauthenticated per the PR:N vector, with no write primitive and no code execution; EPSS is 0.13% (2nd percentile), the issue is not in CISA KEV, and no public exploit was identified at time of analysis though a public technical write-up has been published and an upstream fix commit exists.
Heap out-of-bounds read in libpff (libyal) versions 0 through 20260926 allows a crafted PST, OST or PAB mail-store file to cause the parsing application to copy adjacent heap memory into table record entry values or to crash. The flaw sits in libpff_table_read_values_array, where a 7c-type column definition's values_array_size is used without being checked against the bytes remaining in the values-array buffer. The attack is local and requires the victim to open or otherwise process a malicious file (AV:L, UI:R, PR:N), so impact is bounded to limited heap information disclosure plus availability loss from a crash; EPSS is 0.13% (2nd percentile), there is no CISA KEV listing, and a public technical write-up/PoC reference exists but is not verified as working exploit code.
Heap out-of-bounds read in libpff through 20260926 can expose adjacent heap memory or crash applications such as pffexport and pypff when they parse a crafted PST, OST, or PAB file. The off-by-one bounds check in libpff_table_read_6c_record_entries is reached only when a user or automated pipeline opens the malicious file (local vector, passive user interaction, no authentication), and impact is limited to low confidentiality loss and high availability disruption. No public exploit identified at time of analysis; EPSS is 0.13% (2nd percentile) and the issue is rated moderate rather than emergency priority, with an upstream fix commit available.
Improper authorization in DeepSeek Harness 0.2.0-rc.0 through 0.2.0-rc.2 allows unauthenticated remote attackers to perform unauthorized actions through the Browser Auth component, with low impact to confidentiality, integrity and availability (CVSS 4.0 score 6.9). The flaw resides in packages/client/connection/src/browser-auth.ts and is confined to these pre-release builds; the vendor did not respond to the disclosure and no fixed version is listed. No public exploit code has been identified at time of analysis, and the priority is this_month rather than emergency.
netty-tcnative through 2.0.83.Final passes the raw SNI server_name from the ClientHello into JNI's NewStringUTF(), which decodes modified UTF-8 rather than standard UTF-8. A remote, unauthenticated client can therefore place the overlong two-byte sequence 0xC0 0x80 in the SNI extension and have it materialize as an embedded U+0000 character inside the Java hostname string handed to SniHostNameMatcher and returned by SSL.getSniHostname(), letting the attacker influence SNI-based certificate selection, access-control decisions and logging. The assessed severity is CVSS 3.1 6.5 (AV:N/AC:L/PR:N/UI:N/S:U/C:L/I:L/A:N); the library-level corruption is unconditional, but real-world impact is gated on the integrating application actually relying on that hostname for a trust decision or re-encoding it downstream. The advisory documents a deterministic researcher reproduction against an unmodified build; no public exploit code is referenced, and there is no KEV or EPSS signal in the provided data.
Clickjacking in the boneIO Black web control panel lets a remote attacker frame the interface on a malicious page and trick a user into clicking overlaid controls that switch relays, open gates, or move blinds, producing unauthorized physical state changes. All versions below 1.6.0 are affected per the vendor CPE, and the vendor-released patch is version 1.6.0. Exploitation requires the victim to visit the attacker's page and click a disguised control (UI:A) while the panel is network-reachable; because CVE-2026-77053 leaves authentication disabled by default, any host on the same internal network can be positioned as that victim. This is a genuine but moderate-priority issue, not an emergency: no public exploit code has been identified at time of analysis and there is no KEV listing, with CERT-PL reporting and the vendor documentation as the available sources.
Cleartext credential exposure in boneIO Black before 1.6.0 allows retrieval of MQTT broker and web panel passwords from the GET /api/config endpoint. The assessed vectors carry PR:L, so an authenticated session is normally required, but the reported default configuration tied to CVE-2026-77053 disables authentication, letting an attacker on the same local network obtain the passwords without logging in. Version 1.6.0 fixes the storage and API exposure issues; no public exploit code was identified at time of analysis.
Server-side request forgery in the Erlang hackney HTTP client (versions 4.0.1 up to but not including 4.9.0) lets an attacker steer outbound requests to loopback or private addresses that the calling application believes its SSRF allowlist has already rejected. Two host spellings slip past the check - a percent-encoded bracketed IPv6 literal such as %5B%3A%3Affff%3A127%2E0%2E0%2E1%5D (decoding to [::ffff:127.0.0.1]) and an IP literal with a trailing dot such as 127.0.0.1. - because the caller's validation parses them differently from hackney's connect path; the same applies to a redirect Location target when redirects are followed. Exploitation is conditional on a consuming application that builds its allowlist with uri_string:parse/1 and inet_parse_address/1 and then passes an attacker-influenced URL to a vulnerable hackney release, so attack complexity is high and the CNA scores the issue 6.9 (Medium) under CVSS 4.0; no public exploit has been identified at time of analysis.
Comment impersonation in MacCMS v10 through 2026.1000.4056 allows unauthenticated visitors to attribute anonymous comments to any registered member by submitting a crafted user_id parameter, so the comment renders with that member's avatar and identity. The impact is integrity-only: it does not provide account takeover, data disclosure, or code execution, and it matters mainly where anonymous commenting is enabled and displayed commenter identity is treated as trustworthy. No public exploit was identified at time of analysis.
Authenticated users holding theme management rights on Webasyst Framework through 4.4.0 can read arbitrary files from the server by uploading a crafted theme archive containing symlinks, because waArchiveTar::writeItem() recreates those links without validating their targets and the linked file is then retrievable through the public theme path. Successful abuse exposes secrets such as wa-config/db.php or system files like /etc/passwd. The flaw is gated behind a high-privilege administrative capability (CVSS 3.1 PR:H, 6.9; CWE-59), so realistic attackers are malicious insiders or intruders who already control a privileged account; no public exploit code was identified at time of analysis, and this is tracked as a this-month priority rather than an emergency.