2026-11-03
A use-after-free in Zephyr's MCTP over USB device class lets a physically attached USB host race class teardown against the RX parsing worker, freeing the packet buffer at usb->rx_pkt while the worker still writes host-supplied bytes through it. Only builds of Zephyr 4.4.0 through 4.4.x (fixed in 4.5.0) that enable the experimental CONFIG_MCTP_USB binding are affected; the host needs no credentials or user interaction but must control the bulk OUT endpoint and the SET_CONFIGURATION/bus-reset timing. A public PoC reference exists and was reproduced on native_sim, but it is not verified as working exploit code on real USB hardware, natural-hardware trigger probability remains unverified, and no active exploitation is confirmed.
HDF5 through 2.2.0, when compiled with szip support, reads past the end of a chunk buffer while decoding a szip-compressed chunk in H5Z__filter_szip(): the function subtracts the 4-byte uncompressed-length header from the chunk length without first verifying the chunk is longer than that header, so a truncated or corrupted chunk record makes the size_t subtraction wrap to a very large value and the decompressor reads far beyond the stored data, crashing the process. An attacker triggers it by getting a user or application to read a dataset from a crafted HDF5 file, so the attack is local (AV:L) and requires the victim to open the file (UI:R / UI:P), with impact confined to denial of service (assessed CVSS 3.1 AV:L/AC:L/PR:N/UI:R/S:U/C:N/I:N/A:H; CVSS 4.0 4.7). Only szip-enabled builds (libaec or the szip library) are affected, the CVE is not in CISA KEV, and the CVSS 4.0 vector records Proof-of-Concept-level exploit maturity (E:P) while no verified working exploit code is identified in the available sources; the issue is rated medium severity and prioritized 'this_month' rather than as an emergency.
Zephyr 4.4.0 through 4.4.x (fixed in 4.5.0) drops a Bluetooth Classic AVRCP Target connection without cancelling its vendor-dependent response transmit work or releasing the response buffers still parked on the pending list, so a partly transmitted fragmented response is left behind with the TX_ONGOING flag set. Because the Target objects are static and reused by connection index, the next peer landing on that index inherits a permanently stalled TX machine: every vendor-dependent response queues behind the stale head and never leaves the device, one buffer leaks per cycle from a shallow pool, and a new peer that issues REQUEST_CONTINUING_RESPONSE can be sent the un-transmitted remainder of the previous peer's response. The attack is deterministic but requires an already-paired peer within radio range because the AVCTP control channel mandates an encrypted BR/EDR link; no user interaction is needed, the severity is moderate (CVSS 4.6), and no public exploit code was identified at time of analysis.
Non-terminating symbolic-reference resolution in isomorphic-git 1.38.5 and earlier lets a malicious or man-in-the-middle-controlled Git remote brick a client repository: circular symref= capabilities advertised by the server are persisted to disk by updateRemoteRefs, and any later operation that resolves refs/remotes/origin/a or refs/remotes/origin/b (log, status, checkout, branch) hangs indefinitely. The attack is gated on the victim choosing to clone from the untrusted or compromised remote (UI:R, PR:N), and the result is a persistent denial of service only - no data exposure or code execution (CVSS 3.1 base 4.3, A:L). No CISA KEV entry or confirmed in-the-wild exploitation is reported; the GHSA advisory references a public proof-of-concept repository demonstrating the hang, which is a PoC reference rather than exploit code independently verified as working.
Unbounded memory allocation in isomorphic-git before 1.38.6 allows a malicious or man-in-the-middle-controlled Git remote to force git.clone() or git.fetch() to allocate roughly 2 GB from a single packfile delta, causing denial of service on memory-constrained hosts. The server-controlled delta targetSize is passed to Buffer.alloc without an upper bound in src/utils/applyDelta.js, and Node.js heap limits do not mitigate the off-heap allocation. The vendor fixed this in 1.38.6; a public PoC reference exists but is not verified as working exploit code, and no CISA KEV active-exploitation confirmation is provided.
CVE-2026-97072 is a missing authorization flaw (CWE-862) in the myCred WordPress plugin affecting all versions up to and including 4.0.2, allowing an attacker holding a low-privilege authenticated account to perform actions that should require higher access, with integrity-only impact. The CVSS vector (AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:L/A:N, 4.3) confirms exploitation requires a valid low-privilege account on the target site and is not remote-unauthenticated; no public exploit code or CISA KEV entry is associated with this issue. Risk is moderate and the priority signal points to patching in the normal cycle, since a valid WordPress account on a site running the affected plugin is a prerequisite and no confidentiality or availability impact is assessed.
The myCred WordPress plugin (all versions through 4.0.2) fails to enforce authorization checks on at least one of its actions, letting a logged-in low-privilege user such as a subscriber or customer trigger operations that should require higher capabilities. The impact is integrity-only - site data or myCred point balances and settings may be altered - with no confidentiality or availability impact and no user interaction required beyond having an account. No public exploit code has been identified at time of analysis, and the issue is rated low-to-moderate priority for remediation within the current monthly cycle.
Missing authorization in the WC Lovers WCFM Membership WordPress plugin (wc-multivendor-membership) through 2.12.0 lets a low-privilege authenticated user - for example a registered customer or vendor on a WCFM-powered marketplace - perform actions that should require higher privileges, producing an integrity-only impact (CVSS 4.3, PR:L, no confidentiality or availability loss). Exploitation requires a valid account on the target site and is reachable over the network without user interaction; the specific plugin feature, action or endpoint lacking the check is not described in the available sources. There is no KEV entry and no public exploit identified at time of analysis, and the priority is 'this_month' rather than emergency - a genuine but low-urgency issue.
Script-related HTML injection in the WPMU DEV Forminator plugin for WordPress, in versions up to and including 1.58.3, allows a low-privileged authenticated user to place script-bearing markup into a Forminator feature that accepts input, where the plugin fails to neutralize it. Successful abuse depends on a second party - typically an administrator or another site user - subsequently rendering the injected content, and the scope change in the assessed vector reflects that the payload executes in the victim's browser context rather than the attacker's. No public exploit code and no CISA KEV entry were identified at time of analysis, and this is a low-to-moderate priority issue rather than an emergency.
Information disclosure in Elasticsearch lets an authenticated user whose role applies field-level security (FLS) read the value of a constant_keyword field that the role is supposed to exclude, because certain search and query features do not enforce the FLS restriction for that field type. Only clusters where FLS is configured to exclude a constant_keyword field are exposed, and the leaked data is a single constant value shared across the index, so confidentiality impact is low with no integrity or availability effect; the underlying flaw is an incorrect authorization check (CWE-863 / CAPEC-1). The issue is addressed by the Elastic security update ESA-2026-209 covering versions 8.19.24, 9.4.9 and 9.5.6, and no public exploit identified at time of analysis; priority is assessed as this_month rather than immediate action.
Information disclosure in Elastic Kibana lets an authenticated user who holds read access to connectors in one Kibana space retrieve connector execution log records from other spaces they are not authorized to view. The exposed data is metadata only: connector names and identifiers, execution status, timestamps, error messages and aggregated execution statistics, with no integrity or availability impact (CVSS 4.3, CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:L/I:N/A:N). Exploitation requires a valid low-privilege account and a deployment that uses multiple spaces for access segregation; no public exploit or PoC has been identified at time of analysis, and the issue is not in CISA KEV.
Information disclosure in Elastic Kibana's Machine Learning feature allows an authenticated user with Machine Learning read access in one Kibana space to read audit and error messages for data frame analytics jobs in other spaces, exposing destination index names, analyzed field names, and failure details. The issue is a low-to-moderate severity authorization bypass (CVSS 4.3) limited to job metadata, with no integrity or availability impact, and no public exploit identified at time of analysis. Elastic has released fixes in Kibana 8.19.24, 9.4.9, and 9.5.6.
Elastic APM Server can hand an authenticated APM agent the central agent configuration belonging to a different service - including sampling settings - because of a duplicate-identifier flaw (CWE-694) in how agent configuration requests are resolved. The mix-up lasts only until the cached configuration expires, so the result is timing-dependent rather than reliably attacker-controlled, and it only manifests where one APM Server provides central configuration to more than one service. Exploitation requires agent credentials (CVSS PR:L, network-reachable) and yields limited confidentiality impact; no public exploit has been identified at time of analysis, and Elastic has published fixed releases 8.19.24, 9.4.9 and 9.5.6.
Authenticated users with contributor-level access or above on WordPress sites running the Academy LMS plugin through version 4.0.1 can read the contents of arbitrary .html files from the server by supplying directory-traversal sequences in the 'id' parameter of a REST endpoint; the file contents are returned in full in the JSON response's content field. Impact is deliberately narrow: the plugin appends a '.html' extension to the attacker-supplied path before reading, so wp-config.php, PHP source, and any non-.html file cannot be reached. This is a moderate, not urgent, issue - no active exploitation is confirmed (not in CISA KEV) and no public exploit identified at time of analysis.
Elasticsearch can leak field-level-security-protected information and return incomplete results to other users because cached query results are shared across users with different field-level security (FLS) restrictions for certain query types. Only authenticated users whose access is constrained by field-level security are able to exploit this (CVSS 4.2, AV:N/AC:H/PR:L/UI:N/S:U/C:L/I:L/A:N), and disclosure requires that a full-access user recently ran the identical query on the same shard so that its results are still cached; a restricted user can then infer which documents match a value in a field they cannot read, and can also cause full-access users to receive incomplete search results. Elastic has published a security update (ESA-2026-210) covering fixed releases in the 8.19, 9.4, 9.5 and 9.6 lines; no public exploit code or active exploitation is indicated in the available data, and the issue is assessed as low priority (vuln.today decision: this_month).
The pdfread and nuke plugins shipped with nnn through version 5.3 write to predictable, name- and PID-derived paths under ${TMPDIR:-/tmp}, letting another local user on the same host pre-create a symlink there so the victim's plugin run truncates and overwrites any file the victim can write. Exploitation requires an authenticated local account, write access to the temp directory, and the victim actively invoking the plugin (for the nuke listimages path the attacker must also predict the victim's PID), which confines realistic exposure to shared multi-user systems rather than single-user desktops. There is no remote vector and no public exploit code identified at time of analysis; an upstream commit fixing the plugin temp-file handling is available.