Type confusion in V8 (Chrome's JavaScript engine) in versions prior to 152.0.7977.82 enables unauthenticated remote attackers to execute arbitrary code within Chrome's renderer sandbox by enticing a user to visit a specially crafted HTML page. The flaw exploits V8's type system, allowing an attacker to manipulate object type metadata in the JavaScript heap to gain controlled memory access and code execution inside the renderer process. No public exploit or CISA KEV listing has been identified at time of analysis, though V8 type confusion vulnerabilities in Chrome have historically attracted rapid weaponization given the browser's enormous attack surface and attacker interest.
OS command injection in D-Link DNS-320 ShareCenter firmware 2.06B01 allows a remote authenticated attacker to execute arbitrary operating system commands by supplying a crafted value for the fileurl parameter in the File Sharing CGI handler (/cgi/file_sharing.cgi). The vulnerability carries a CVSS 4.0 base score of 8.5 with high-privilege requirement but yields full system-level compromise on success, including subsequent-system impact. A public exploit has been disclosed, materially lowering the barrier to exploitation for any remaining deployments of this legacy NAS appliance.
OS command injection in D-Link DNS-340L firmware 1.01B04 allows authenticated remote attackers to execute arbitrary operating system commands by manipulating the `callback_url` or `sync_interval` arguments passed to the `/cgi-bin/dropbox.cgi` CGI handler. A public proof-of-concept exploit exists, raising the likelihood of opportunistic exploitation against exposed NAS devices. No patch has been identified at time of analysis, and the DNS-340L is likely an end-of-life product with no vendor remediation roadmap.
Pre-authentication remote buffer overflow in Tenda HG10 router firmware version 300001138 allows unauthenticated network attackers to execute arbitrary code by sending an oversized Username value to the Boa Web Server's /boaform/formLogin endpoint. The formLogin function performs no bounds checking on the supplied credential field, enabling stack or heap corruption reachable directly from the WAN or LAN interface without any prior authentication. A public proof-of-concept exploit has been disclosed on GitHub, materially lowering the bar for exploitation. No public exploit identified at time of analysis is overridden here: publicly available exploit code exists.
Remote code execution in NousResearch Hermes Agent 0.18.2-0.21.0 allows an unauthenticated attacker to execute arbitrary OS commands in the victim's process context by supplying a malicious git repository containing a crafted `.git/config` that sets `core.fsmonitor` to an attacker-controlled binary or script. Exploitation is triggered the moment a user opens the malicious repository and sends any message, causing the agent's internal git status call to invoke the injected command and expose the full process environment - including configured LLM provider API keys. A public exploit writeup exists (manifold.security), and a patch was released as commit f6234d0.
Unauthenticated SQL injection in SciPhi-AI R2R through version 3.6.6 exposes the application database to blind data exfiltration via the retrieval search endpoint's filter key parameter. The root cause is direct string interpolation of attacker-controlled filter keys into SQL WHERE clauses with no parameterization or escaping, enabling both time-based and boolean-based extraction techniques. A publicly available proof-of-concept exploit exists (VulnCheck, GitHub issue #2308), materially elevating the practical risk beyond the theoretical; no CISA KEV listing at time of analysis.
Unauthenticated arbitrary file upload in GeoNetwork prior to versions 4.4.12 and 4.2.17 allows remote attackers to write malicious XSLT or ZIP formatter files directly to the server's formatter directory without any authorization check. Because GeoNetwork processes uploaded `.xsl` formatter files server-side via a Java XSLT engine when rendering metadata records, attackers can weaponize this upload into a remote code execution chain by embedding Java extension function calls within the XSLT payload. No public exploit has been confirmed at time of analysis, but The Hacker News coverage explicitly frames this as unauthenticated RCE, and the exploitation path (malicious XSLT to Java extension-function execution) is a well-documented Java attack technique.
Heap buffer overflow in zlib 1.3.1.2 through 1.3.2 allows remote attackers to corrupt memory via crafted input to applications using the non-blocking gzwrite API. The gz_vacate() function fails to bounds-check a memmove() when stale external buffer pointers are present after a write stall, reachable by calling gzprintf() or gzvprintf() in that state. A publicly available proof-of-concept exploit exists; no vendor-released patch has been identified at time of analysis.
Prototype pollution in toml-node (npm package `toml` by BinaryMuse) prior to 4.1.2 allows unauthenticated remote attackers to write arbitrary keys onto Object.prototype by supplying a crafted TOML document to `toml.parse()`. The vulnerability exploits a path desynchronization bug in the compiler's duplicate-key guard - comma-joined tracking paths diverge from dot-joined traversal paths, allowing `__proto__` traversal through scalar values into the real JavaScript prototype chain. Injected properties propagate process-wide in Node.js and can enable denial of service, authorization bypass, or code execution via application gadget chains; the GHSA advisory includes working proof-of-concept code, and a fix is available in version 4.1.2.
Uncontrolled recursion in toml-node (npm package `toml`, ~47 million monthly downloads) allows remote unauthenticated attackers to crash any Node.js process that parses attacker-controlled TOML input. The Peggy 5.1.0-generated recursive-descent parser has no depth limit in the mutual recursion between `peg$parsevalue`, `peg$parsearray`, and `peg$parseinline_table_entry`; a ~5-6 KB payload of deeply nested arrays reliably exhausts the call stack and raises an uncatchable `RangeError` that terminates the request worker or process. A confirmed PoC is documented in the GHSA advisory; a patch is available in version 4.2.0 released July 13, 2026.
Buffer overflow in Tenda HG10 router firmware version 300001138 enables network-accessible authenticated attackers to overflow memory in the Boa embedded web server's WLAN configuration handler, potentially achieving arbitrary code execution. The vulnerable parameter is `ssid` within the `/boaform/formWlanSetup` endpoint, which lacks input length validation per CWE-120. A public proof-of-concept is hosted on GitHub, concretely lowering the exploitation bar; no vendor patch has been identified at time of analysis.
Unauthenticated UDP packet injection in MOOS essential-moos pMOOSBridge through version 10.0.1 allows any network-reachable attacker to inject arbitrary variables into the local MOOS publish-subscribe community by sending crafted UDP datagrams to the configured UDPListen port, with full control over spoofed source and community identifiers. The root cause is the absence of any source-address validation in the ListenLoop: the bridge historically accepted datagrams from any sender, trusting the identity claims embedded in the payload. No public exploit has been identified at time of analysis, and a fix is available via upstream PR/commit, though a tagged release version has not been independently confirmed.
Unauthenticated UDP datagram injection in essential-moos pShare through 10.0.1 allows any network-reachable attacker to send crafted UDP packets that pShare republishes verbatim to the local MOOS community under an attacker-chosen identity, or to crash the pShare bridge process with malformed payloads. The root flaw is that pShare's Listener class used raw socket reads without capturing the sender's address, performing no source validation before deserializing and forwarding messages as authoritative MOOS community traffic. A fix is available in upstream commit 53729b6 and PR #18, which adds allowlist-based source filtering and wraps deserialization in exception handling; no public exploit code has been identified and the CVE is not in CISA KEV.
Heap corruption in MOOS essential-moos through 10.0.1 allows unauthenticated remote attackers to crash the pMOOSBridge process by sending a single crafted UDP datagram containing a negative declared packet length. The root cause is a signed-to-unsigned type confusion in CMOOSUDPLink::ReadPktFromArray(): a negative nRqd value passes a signed bounds check unchanged, then is silently cast to a near-SIZE_MAX size_t when passed to memcpy, writing far beyond the destination heap buffer. No public exploit has been identified at time of analysis, but the attack primitive requires only network access to the UDP listen port and a small crafted payload; the upstream fix is available as a GitHub patch commit.
OS command injection in the D-Link DNS-340L NAS firmware 1.01B04 allows a remote administrator to execute arbitrary operating system commands via unsanitized parameters in the Add-On Center CGI handler at `/cgi-bin/addon_center.cgi`. The parameters `f_name`, `f_url`, `f_flag`, and `f_login_user` are passed without sanitization to a shell command, enabling full root-level device takeover. A public proof-of-concept exploit has been disclosed on GitHub, raising practical exploitability beyond the theoretical; no vendor patch has been identified, and the DNS-340L is an end-of-life device unlikely to receive one.
Authorization bypass in Eclipse Arrowhead 5.0.0-5.2.1 allows any authenticated system - regardless of privilege level - to reach every protected management API endpoint by submitting percent-encoded paths that defeat the raw-URL substring check while still being routed by Spring MVC's decoded dispatcher. Exploiting the flaw is trivial: encoding a single character in '/mgmt/' (e.g. 'm' → '%6D') causes the filter to fall through without authorizing the request, yet Tomcat decodes and dispatches it normally. The highest-impact target is POST /authentication/mgmt/identities, which creates new sysop-level accounts, yielding full administrative takeover of the Arrowhead local cloud. No public exploit has been identified at time of analysis.
Authorization bypass via user-controlled key in Azure Cosmos DB enables an authenticated attacker to perform cross-account spoofing over the network, potentially accessing or manipulating data belonging to other accounts or tenants. Exploiting CWE-639 (IDOR-class flaw), an attacker with a valid Cosmos DB account can supply a crafted key to circumvent per-resource authorization checks, yielding full confidentiality, integrity, and availability impact per the CVSS vector. No public exploit code has been identified and SSVC confirms no active exploitation, though CVSS impact metrics indicate total technical impact if successfully triggered.
Privilege escalation in Microsoft Fabric allows an authenticated low-privileged attacker to elevate their access level over the network, gaining full control over confidentiality, integrity, and availability of affected resources. The CVSS 3.1 vector (AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) confirms low-friction network exploitation requiring only a valid user account and no additional interaction. No public exploit code has been identified and CISA has not listed this in KEV, though SSVC rates the technical impact as total.
Unauthenticated buffer over-read in MOOS core-moos through 10.4.0 allows any attacker with TCP access to the MOOSDB port to crash the server or leak process memory by sending a crafted 4-byte packet. The CMOOSCommPkt deserialization path fails to validate that the declared packet length meets the minimum header size before calling Serialize(), causing it to read the message-count field from memory beyond the end of the received data. No authentication is required - exploitation precedes any credential exchange, making this reachable on any exposed MOOSDB instance. No public exploit has been identified at time of analysis, but the fix PR (GitHub #75) precisely documents the trigger condition.
Type confusion (CWE-843) in the Compositing subsystem of Google Chrome prior to 152.0.7977.82 enables unauthenticated remote attackers to execute arbitrary code within the renderer sandbox by delivering a crafted HTML page to a victim. Execution is confined to Chrome's renderer process rather than the underlying OS - a full host compromise would require chaining with a separate sandbox escape primitive. No public exploit or active exploitation (KEV) has been identified at time of analysis; SSVC rates Exploitation: none and Automatable: no.
Use-after-free in Skia, Chrome's 2D graphics rendering engine, allows remote attackers to execute arbitrary code within the Chrome renderer sandbox by luring a user to a crafted HTML page. Affected versions are all Chrome releases prior to 152.0.7977.82 across desktop platforms. The sandbox confinement means this vulnerability alone does not yield full system compromise - a chained sandbox escape would be required for that - but the renderer-process code execution is rated high impact by Chromium's internal severity team. No public exploit or CISA KEV listing is present at time of analysis; SSVC labels exploitation as none and the attack as non-automatable due to the required user interaction.
Sandboxed remote code execution in Google Chrome's CacheStorage API (prior to 152.0.7977.82) allows a remote attacker to execute arbitrary code inside the Chrome renderer sandbox by luring a user to visit a specially crafted HTML page. The root cause is CWE-668 (Exposure of Resource to Wrong Sphere), where CacheStorage improperly exposes resources to an unintended security context. No public exploit code has been identified and SSVC reports exploitation status as none, though the patch is available and update is strongly recommended.
MOOSDB in core-moos through 10.4.0 accepts wire-supplied source identifiers verbatim instead of enforcing the authenticated connection identity, enabling any authenticated MOOS client to forge message origins and impersonate arbitrary registered processes within the broker. Exploitation allows an attacker to attribute writes to a victim client and invoke subscription cancellations as if originating from that client, corrupting the publish-subscribe integrity of the entire MOOS community. No active exploitation has been confirmed (not in CISA KEV), but the vulnerable code path is fully exposed by the published fix in GitHub PR #76, making the flaw straightforward to exploit for any connected attacker.
CSRF in MISP's cullEmptyEvents endpoint allows a network attacker to trigger irreversible mass deletion of published empty events by causing an authenticated, privileged MISP user's browser to auto-request a crafted URL - for example via an embedded image tag. The endpoint accepted HTTP GET requests, which CakePHP's SecurityComponent does not CSRF-validate because GET requests carry no body, causing the framework's hasData check to return false and bypass token enforcement. The deletion is particularly destructive because it runs with skipBlocklist set, meaning purged events leave no blocklist entries to prevent or track subsequent re-synchronization; fix commits are available on GitHub, though the exact patched release version is not independently confirmed. No public exploit code is identified at time of analysis.
Btrfs in the Linux kernel fails to reinitialize inode address-space mapping flags and minimum folio order when re-reading previously evicted cached inodes, causing a kernel assertion failure and system crash (denial of service). The defect surfaces only when btrfs is mounted with a block size larger than the host memory page size - a non-default but valid configuration - and manifests after a create/evict/re-read inode cycle. No public exploit code exists and EPSS is 0.20% (10th percentile), reflecting negligible real-world exploitation probability beyond accidental triggering.
SAML response signature validation is absent or insufficient in the Mendix SAML module across all Mendix 9.24, 10, and 11 compatible releases prior to their respective patches, enabling unauthenticated remote attackers to forge SAML assertions and hijack arbitrary user sessions under specific SSO configurations. The flaw (CWE-347) is conditional on deployment configuration and requires passive user participation in an active SSO flow, moderating its exploitability from a mass-opportunistic standpoint. Vendor-released patches are available; no public exploit code or CISA KEV listing has been identified at time of analysis.
Command injection in SEPPmail Secure Email Gateway before version 15.0.7 enables authenticated administrators to execute operating system commands at elevated privilege levels beyond the intended administrative scope. Classified CWE-78, the flaw exists within the network-accessible management interface and was reported by NCSC.ch (Swiss National Cybersecurity Centre). No active exploitation or public proof-of-concept code has been identified at time of analysis; the vendor-released fix is available in version 15.0.7.
Remote denial of service in OpenVPN's reliability layer - the custom reliable-delivery sublayer underpinning the TLS control channel - allows unauthenticated network attackers to crash or hang OpenVPN daemons on versions up to 2.7.6 and 2.6.22 by triggering an unbounded TLS retransmission timeout or forcing acknowledgment of packets that cannot be outstanding. The CVSS 4.0 vector (AV:N/AC:L/AT:N/PR:N/UI:N/VA:H) confirms full availability impact with no authentication, complexity, or user-interaction requirements, and downstream VPN clients lose connectivity when the concentrator is taken offline. No public exploit code or active exploitation (KEV) has been identified at time of analysis; vendor-released fix is available as OpenVPN v2.7.7.
PHP Object Injection in the RTMKit WordPress plugin (versions ≤ 2.1.5) allows an authenticated Contributor-level user to pass a crafted serialized PHP payload through a vulnerable plugin input, triggering deserialization of untrusted data. In plugin-rich WordPress environments where PHP gadget chains are commonly available, this escalates to remote code execution, full database access, and file system compromise. No confirmed active exploitation or public exploit code has been identified at time of analysis; the vulnerability was disclosed by Patchstack.
Path traversal in Eclipse aeriOS Self-orchestrator prior to version 1.2.1 allows unauthenticated remote attackers to write or delete arbitrary JSON files on the host filesystem by embedding path separator sequences (e.g., '../') in resource identifiers submitted to the REST API. The vulnerability stems from direct string concatenation of user-supplied names into `fs.writeFileSync()` paths without sanitization, confirmed by the upstream commit diff. Real-world impact is amplified by the complete absence of API authentication (PR:N) and by the container running with elevated privileges in the affected deployment, extending the writable filesystem scope beyond the intended `./facts/` directory. No public exploit code has been identified at time of analysis.
Cart-record tampering in J2Store (Joomla e-commerce extension by j2commerce.com) affects all versions across the 1.0.0-3.3.21, 4.0.0-4.0.21, and 4.1.0-4.1.6 release lines via an unoverridden inherited FOF framework `save` task reachable by unauthenticated front-end requests. The controller correctly restricted `remove()` to the caller's own session but never overrode the generic FOF `save` task, leaving it exposed to unauthenticated `format=raw` POST requests that can insert new cart rows with attacker-chosen `user_id`/`session_id` values or overwrite existing rows by numeric ID. No public exploit has been identified at time of analysis, but the zero-authentication, low-complexity network vector makes this trivially exploitable against any publicly reachable J2Store storefront.
Unbounded decompression of gzip and ZIP attachments in parsedmarc before 11.0.1 exposes any organization running automated DMARC report processing to remote, unauthenticated denial of service. Because `extract_report()` inflates compressed attachments with no size ceiling, a decompression bomb exploiting deflate's ~1000:1 ratio on degenerate input - e.g., a 100 KB attachment expanding to ~100 MB in RAM, with peak usage near 200 MB - exhausts host memory when parsedmarc auto-processes the email. No public exploit or active exploitation has been identified at time of analysis, but the attack surface is trivially reachable by any sender whose emails are accepted by the monitored mailbox.
Uncontrolled memory allocation in MOOS core-moos through 10.4.0 exposes MOOSDB servers to pre-authentication denial of service. The CMOOSCommPkt::OnBytesWritten() function reads the packet length field directly from untrusted wire data and passes it without any upper-bound check to new unsigned char[], allowing unbounded heap allocation before any client authentication handshake occurs. An unauthenticated network attacker can craft packets with arbitrarily large declared lengths to exhaust server memory and crash the MOOSDB process; no public exploit has been identified at time of analysis, and an upstream patch is available via GitHub commit e7ea624.
Unbounded memory consumption in MOOS-IvP pMarineViewer through version 24.8.1 allows unauthenticated network attackers to exhaust the operator display process by flooding it with NODE_REPORT messages carrying arbitrary distinct node names. The ContactLedger class allocates a new tracking entry for every unique vessel name it encounters without enforcing any cap, making the ledger - and everything that iterates it - grow without limit. The result is a stalled or crashed operator situational-awareness display, which is operationally significant in autonomous maritime missions. No public exploit has been identified at time of analysis; an upstream fix capping contacts at 200 has been merged but a tagged release version has not been independently confirmed.
Prototype chain sandbox escape in n8n's $fromAI expression handler allows any user holding workflow-build privilege to achieve remote code execution within the main n8n server process. The $fromAI handler failed to restrict placeholder key lookup to own properties and permitted reserved prototype keys; when the input value is a primitive, the handler returns a live reference to the host-prototype chain, enabling traversal to Function.prototype.constructor and compilation of arbitrary JavaScript outside the expression sandbox. All n8n deployments running versions below 1.123.73 (v1 track), 2.35.4, or 2.36.2 (v2 track) are affected. No public exploit code or CISA KEV listing has been identified at time of analysis, but the technique (prototype chain walk to Function constructor) is well-understood and reliable once sandbox access is obtained.
Path traversal in rubyzip before 3.4.0 allows an attacker who supplies a crafted ZIP archive to write files outside the intended extraction directory by exploiting a flawed prefix comparison in `Zip::Entry#extract`. The root cause is that the path guard checks `extract_path.start_with?(dest_dir)` without appending a trailing file separator, meaning a sibling directory sharing the destination string prefix - such as `out_evil` alongside destination `out` - passes validation and receives the extracted file. The fix is confirmed in v3.4.0; no active exploitation has been identified at time of analysis, but the attack primitive is straightforward to reproduce from the published commit diff.
Stack buffer overflow in the uMS graphical operator tool of MOOS ui-moos through commit 50b9c6c allows an unauthenticated network attacker to achieve code execution on an operator workstation by injecting excessively long client or variable names into the MOOSDB pub-sub middleware. When the operator subsequently selects a malicious process list entry or pokes an oversized variable through the uMS interface, unchecked sprintf calls in ScopeTabPane.cpp and ScopeGrid.cpp overflow fixed 1024-byte stack buffers, enabling memory corruption and control-flow hijack. The upstream fix (PR #5, commit a6ebc0bc) is confirmed by direct patch diff evidence replacing all three vulnerable sprintf calls with snprintf; no public exploit code or KEV listing has been identified at time of analysis.
Unauthenticated denial-of-service in MOOS core-moos through 10.4.0 allows remote attackers to crash the central MOOSDB process by sending a single crafted message containing a negative serialized string length value. The negative integer passes the existing arithmetic check in CMOOSMsg::operator>> but is then implicitly widened to an enormous size_t by string allocation routines, triggering an unhandled std::exception that escapes the server accept loop and terminates the MOOSDB process - halting all MOOS inter-process communication. No public exploit code or CISA KEV entry is confirmed, but the VulnCheck advisory and linked source code provide sufficient detail to reproduce the crash trivially.
Resource exhaustion in the MOOSDB HTTP server component of MOOS core-moos through 10.4.0 allows unauthenticated remote attackers to render MOOSDB unavailable by opening many simultaneous connections and sending unbounded HTTP header data. Source code analysis confirms the pre-patch implementation in CMOOSDBHTTPServer and CHTTPConnection imposed no limits on concurrent connections, header line count, per-line byte size, or per-connection read timeouts - all four controls were absent. No public exploit code or active exploitation has been identified at time of analysis, though the attack requires no authentication or special tooling and is trivially automatable.
Unbounded resource allocation in MOOS-IvP uFldShoreBroker through 24.8.1 allows a remote unauthenticated attacker to exhaust the shore broker process's memory and degrade per-iteration CPU performance by sending NODE_BROKER_PING messages with an unlimited number of distinct community names. Each unique community name causes entries to be appended across six parallel in-memory vectors with no eviction policy and no enrollment ceiling, creating unbounded O(n) per-pass overhead. Sustained delivery crashes or disables the shore broker, severing coordination between shore-side operators and autonomous marine vehicles. No public exploit has been identified at time of analysis.
Unauthenticated remote resource exhaustion in MOOS-IvP pRealm through version 24.8.1 allows any network-reachable attacker to register unlimited long-lived REALMCAST_REQ subscriptions with arbitrarily large variable lists, causing the pRealm process to serialize and publish excessive output on every scheduling interval until system resources are exhausted. The vulnerability stems from three absent input limits: no cap on subscription duration, no cap on variable count per subscription, and no ceiling on concurrent pipeway registrations. The fix, confirmed by PR #132 and commit 8a4bb9d9, introduces hard limits of 3600 seconds duration, 64 variables per request, and 64 total concurrent pipeways. No public exploit or KEV listing has been identified at time of analysis.
Quadratic algorithmic complexity in MOOS-IvP's uFldNodeComms shoreside broker (versions through 24.8.1) allows unauthenticated remote attackers to progressively degrade and halt legitimate autonomous vehicle node-report distribution by injecting NODE_REPORT messages with an unbounded stream of distinct fake node names. Each new name triggers an all-pairs distribution walk over the entire ledger (O(n²) work per pass), causing the shoreside broker to consume exponentially increasing CPU as attacker-controlled name diversity grows. No public exploit code has been identified and the vulnerability is absent from CISA KEV, but the attack is mechanically trivial against any exposed MOOS message bus.
Unbounded memory allocation in MOOS-IvP through 24.8.1 allows unauthenticated network attackers to exhaust host system memory by sending crafted mux packets with arbitrarily large declared packet counts to the Demuxer::addMuxPacket() function. The demultiplexer trusts the integer value in the mux header without a ceiling, causing the process to reserve memory for each declared packet slot. No public exploit code has been identified at time of analysis, and no active exploitation is confirmed by CISA KEV.
Unauthenticated network attackers can trigger out-of-bounds memory reads in MOOS-IvP through version 24.8.1 by sending NODE_REPORT messages containing leading or trailing whitespace. The three affected functions - isQuoted(), isBraced(), and isChevroned() in MBUtils.cpp - strip whitespace into a modified string but then index that shorter buffer using the original string's length, causing reads past the end of the stripped buffer and exposing adjacent heap memory or crashing the process. No public exploit or CISA KEV listing has been identified at time of analysis; the upstream fix is available as a GitHub commit and pull request but a tagged release version has not been independently confirmed.
Denial-of-service in MOOS core-moos through 10.4.0 allows any unauthenticated network peer to permanently freeze the MOOSDB accept thread by opening a TCP connection to the MOOSDB port and sending no data during the wire-protocol handshake. Because the accept thread executes a blocking receive without any timeout while holding the socket-list lock, a single stalled connection prevents all subsequent MOOS clients from registering, effectively rendering the publish-subscribe middleware inoperable. No public exploit code or active exploitation has been identified; however, the attack requires only a single TCP connection and no credentials, making it trivially automatable against any exposed MOOSDB instance.
Authentication bypass in CRMEB through v6.0.0 allows any low-privileged authenticated user - including sub-administrators and accounts carrying no assigned roles - to access restricted admin endpoints. The root cause is a structurally broken role-check implementation in the verifyAuth() method of SystemRoleServices.php, where both conditional branches unconditionally return true, rendering the authorization gate completely inert. VulnCheck has published a dedicated advisory and the flaw is documented in a public GitHub issue (#119), but no CISA KEV listing or confirmed active exploitation has been identified at time of analysis.
SiYuan's full-text search logging mechanism writes admin API tokens in plaintext to an accessible log file when requests exceed internal timing thresholds, enabling authenticated low-privilege users to escalate to permanent administrative access. All self-hosted SiYuan instances before v3.8.2 are affected; because API tokens are passed as query parameters and the log file is retrievable via the built-in getFile endpoint, the privilege escalation path requires no additional exploitation steps beyond triggering a slow search and reading the file. No public exploit code has been identified at time of analysis, and the vulnerability is not listed in the CISA KEV catalog. Vendor-released patch v3.8.2 resolves the issue.
Server-Side Request Forgery in Ollama versions 0.30.0 through 0.33.2 allows unauthenticated remote attackers to redirect the Ollama server's blob-download requests to arbitrary internal hosts, including cloud metadata endpoints such as the AWS instance metadata service. The flaw exists in the tensor-layer model pulling code path (x/transfer/download.go), where HTTP redirect destinations are not validated against an allowlist or the originating registry host. With no public exploit identified at time of analysis, the primary real-world impact is credential theft from cloud metadata services when Ollama runs on cloud-hosted infrastructure.
Unauthenticated SQL injection in WWBN AVideo's channels API endpoint allows remote attackers to extract password hashes and account recovery tokens without any credentials. The `sort` column parameter in `get.json.php?APIName=channels` is injected directly into a SQL ORDER BY clause unsanitized, enabling an ordering oracle attack to infer sensitive column values character-by-character and triggering verbose SQL errors that disclose the full query and database schema. No public exploit has been identified at time of analysis, though VulnCheck has independently reported this vulnerability with a GitHub Security Advisory published by the vendor.
Improper escrow accounting in Gardens v2's StreamingEscrow contract allows a proposal beneficiary to receive the full available escrow balance when an approve-side dispute is resolved in their favour, rather than the correct depositAmount() entitlement. Affected commit is dfba919e218e20d52db9f7b2e8d292d45a46c91b and all prior revisions. No patch has been released at time of publication. No public exploit or CISA KEV listing has been identified, but the logic flaw is described in a public GitHub Security Advisory (GHSA-3xpr-2mm7-77j7), lowering the bar for exploitation by anyone familiar with the codebase.