The internet is broken right now.
44 critical CVEs without a fix in the last 48 hours. Developers are crying quietly into their IDEs. 61 more already have a fix that someone still has to apply.
-
CVE-2026-129866
Dinky through 1.2.5 contains a missing authorization vulnerability in ClusterConfigurationController that allows authenticated users without cluster configuration permissions to list Flink cluster configurations. Attackers can query the /api/clusterConfiguration/list and listAll endpoints to retrieve full configJson deployment settings for every cluster in their tenant.
-
CVE-2026-126272
NocoDB before 2026.09.0 contains an uncontrolled resource consumption vulnerability that allows authenticated users to bypass bulk row-count limits by using session authentication instead of API tokens. Attackers with Editor write permission on one table can submit hundreds of thousands of rows to v1 and v2 bulk write routes, exhausting database connections, CPU, and memory.
-
CVE-2026-126616
The auto-ban of Progressive Robot hMailServer 6.2.0 through 6.3.3, with a MySQL or MariaDB database, can be kept from banning an address after repeated failed logons, which allows a remote unauthenticated attacker to go on guessing account passwords from a single address without being banned. The server makes its MySQL tables in MySQL's utf8 character set, which holds no character outside the Basic Multilingual Plane, sends its statements as UTF-8 over a utf8mb4 connection and leaves the database's default strict mode in place. From 6.2.0 it decodes the user name of a SASL PLAIN logon as UTF-8, so the name can hold such a character. When the failed logons counted from one address reach MaxInvalidLogonAttempts (3 by default), the server clears that address's count, saves an IP range named "Auto-ban: " followed by the user name of the logon that reached the limit, and drops the connection. If such a character is in the part of the user name that the range's name keeps (the name is cut to 100 characters), the database refuses the range: no range is saved, the address is not refused, and, its count having already been cleared, it is given the next full set of attempts. An attacker can in this way spend all but the last attempt of each set on password guesses, and make the last under such a name. On Windows, POP3 AUTH PLAIN reaches it on a server as installed, and IMAP AUTHENTICATE PLAIN where IMAP SASL authentication is turned on (off by default); on Windows and on Linux (from 6.3.0), SMTP AUTH PLAIN reaches it where plain-text authentication is allowed (off by default), and ManageSieve and CardDAV where they are turned on (off by default). Each failed save is reported in the server's error log. A database whose strict mode has been turned off stores the character as a question mark and makes the first such ban, but refuses every later ban named after the same user name, or after one holding other such characters in their place, as a duplicate while the first stands, since the server cannot look up a name holding such a character there.
-
CVE-2026-107474
The Apache Airflow connections API did not mask the apprise provider's `config` connection extra, which can embed notification-service credentials. A user with permission to read connections could retrieve it in cleartext, although Airflow 3 treats connection credentials as write-only. Users are advised to upgrade to apache-airflow 3.4.0 or later. Airflow 2.x is also affected; Airflow 2 has reached end of life and will not receive a fix, so Airflow 2 users should upgrade to Apache Airflow 3.4.0 or later.
-
CVE-2026-63120
pghoard is a PostgreSQL backup daemon and restore tooling that stores backup data in cloud object stores. Prior to version 2.7.2, an incomplete path traversal fix in pghoard allows an attacker with write access to the backup object storage (such as S3, GCS, or Azure) to execute arbitrary file writes and directory creation outside the designated PostgreSQL data directory (pgdata). When a database administrator executes a restore operation (via pghoard_restore get-basebackup), a maliciously crafted delta backup manifest with path traversal sequences (e.g., ../../../../etc/cron.d/backdoor) will write arbitrary payloads to the host system. This issue has been patched in version 2.7.2.
-
CVE-2026-125118
Mongoose is a MongoDB object modeling tool designed to work in an asynchronous environment. Prior to 6.13.12, 7.8.13, 8.24.7, and 9.11.4, Schema.prototype.virtual() builds nested virtual paths without applying the special-property validation already used by Schema.path() and Schema.add(). An attacker who can influence virtual field names in dynamic schemas, content-management form builders, or schema-as-code systems can supply path segments such as __proto__, constructor, or prototype, causing the virtual-path reducer to write enumerable properties to Object.prototype or Function.prototype. The resulting prototype pollution can spoof inherited property checks such as authorization flags and can provide function-prototype gadget state to other server-side JavaScript code. This issue is fixed in versions 6.13.12, 7.8.13, 8.24.7, and 9.11.4.
-
CVE-2026-121820
A vulnerability has been identified in armeria-saml versions prior to 1.43.0, where the HTTP-Redirect binding decoder decompresses attacker-supplied SAML messages without limiting the output size before signature validation, allowing an unauthenticated remote attacker to exhaust server memory and cause a denial of service.
-
CVE-2026-104608
A remote code execution vulnerability was identified in GitHub Enterprise Server in the Git service (babeld). babeld held references into internal collections while calling into the SSH library, where a re-entrant callback could modify a collection and invalidate those references. An attacker with an account and SSH access to the instance could leverage this use-after-free into a controlled write and run arbitrary commands as babeld, including disclosure of its environment secrets. Exploitation required per-build memory offsets; failed attempts crashed babeld. This affected all releases prior to 3.23 and was fixed in 3.22.3, 3.21.8, 3.20.10, and 3.19.14. This vulnerability was reported via the GitHub Bug Bounty program.
-
CVE-2026-81559
A missing authorization vulnerability in the TaskScheduler API in Synology DiskStation Manager (DSM) before 7.3.2-86009-5 and 7.4.1-90080-1 allows remote authenticated users to write specific files.
-
CVE-2026-81558
A missing authorization vulnerability in the SocketIo API in Synology DiskStation Manager (DSM) before 7.3.2-86009-5 and 7.4.1-90080-1 allows remote authenticated users to write specific files.
-
CVE-2026-19414
A missing authorization vulnerability in the Promotion API in Synology DiskStation Manager (DSM) before 7.3.2-86009-5 and 7.4.1-90080-1 allows remote authenticated users to write specific files.
-
CVE-2026-105437
Affected versions of Git have a heap buffer overflow in the three-way merge code (xdiff), which an attacker can trigger through a repository they control. When Git merges a file, it first adds up the size of the merged output, allocates a buffer of that size, and then writes the output into it. That total was kept in a 32-bit signed integer. The conflict-marker-size attribute, which can be set from a committed .gitattributes file, accepts any positive integer, and the markers are written several times for every conflict. A small file with a huge marker size and many separate conflicts can therefore push the total past 4 GiB, so the integer wraps to a small value and Git writes attacker-controlled merge output past the end of a buffer that is too small. At minimum this crashes Git, and because the overflow is large and its contents are largely attacker-controlled, worse heap corruption is possible. Any command that does a textual three-way merge can trigger it, such as git-merge(1), git-rebase(1), git-cherry-pick(1), and git-merge-tree(1). The latter is most likely a target for Git forges, which might use git-merge-tree(1) on the server-side to check mergeability. When the server reads attributes from trees in the repository (for example through attr.tree or --attr-source), any user who can push branches and request that they be merged (for example by opening a pull or merge request) can trigger the overflow without anyone else doing anything, and the data they push is small because the oversized output only appears during the merge. The problem has been patched by computing the size as size_t and checking every addition for overflow, so Git stops with an error instead of overflowing. Servers that cannot upgrade right away can add * conflict-marker-size=7 to each repository's info/attributes as a workaround, which overrides any in-tree .gitattributes, or turn off server-side mergeability checks for untrusted repositories.
-
CVE-2026-95839
containerd is an open-source container runtime. From 2.2.0 until 2.2.10, 2.3.7, and 2.4.2, image manifest processing through the transfer service can encounter a child descriptor containing the io.containerd.manifest.subject annotation and a malformed digest. Pulling or importing such an image, including through a default CRI PullImage request, can trigger an unrecovered panic and crash the containerd daemon. In Kubernetes, kubelet retries can repeatedly trigger the crash and disrupt container operations on the worker node. This issue is fixed in versions 2.2.10, 2.3.7, and 2.4.2.
-
CVE-2026-63475
isomorphic-git is a pure JavaScript implementation of Git for Node.js and browsers. Prior to 1.38.6, isomorphic-git persists server-supplied circular symref= capabilities through updateRemoteRefs, and GitRefManager.resolve() follows the resulting loose symbolic references without a default depth limit or cycle detection. After a clone from a malicious or man-in-the-middle-controlled remote, operations such as log, status, checkout, and branch that resolve refs/remotes/origin/a or refs/remotes/origin/b do not terminate. The circular references remain on disk, leaving the repository unusable for the application and causing a persistent denial of service. This issue is fixed in version 1.38.6.
-
CVE-2026-82179
Dell PowerStore contains an authorization bypass vulnerability. An authenticated administrator could exploit this vulnerability to bypass multi-approver controls on critical operations.
-
CVE-2026-127279
A vulnerability was detected in AIP-PUB ACPs-community up to 2.2.0. This impacts an unknown function of the file discovery-server/app/main.py of the component Discovery Server. The manipulation results in missing authentication. The attack may be launched remotely. The exploit is now public and may be used. The vendor was contacted early about this disclosure.
-
CVE-2026-87857
Temporal Server does not bind a workflow-history pagination token to the namespace and workflow named in the request that presents it. GetWorkflowExecutionHistory and GetWorkflowExecutionHistoryReverse authorize the caller against the namespace in the request, then select history from the branch recorded inside the caller-supplied next_page_token. The normal history continuation carries a run identifier, event bounds, a persistence cursor, and a branch token, but no namespace or workflow identity, and it is serialized as unauthenticated protobuf bytes. An authenticated caller holding read permission on one namespace, together with a genuine non-final continuation issued earlier for a workflow in a second namespace, can submit that continuation with a request naming the first namespace and receive the second namespace's workflow history, provided the supplied workflow identifier selects the same History shard as the continuation's branch. No workflow with the supplied identifier needs to exist. Because the token is unauthenticated, a holder can also advance its event range and clear its persistence cursor, which returns history events created after that caller's current credential no longer granted access to the continuation's namespace. Determining exposure is a version question rather than a configuration question: every release listed as affected omits the binding check on the forward history path, and on the reverse history path in releases from 1.16.0, which introduced GetWorkflowExecutionHistoryReverse; no setting in an affected release restores it, and affected releases emit no metric or log for the condition, so their telemetry provides no detection. Fixed releases add the check and, when a page token's branch token is not the requested execution's current branch token, record the counter pagination_token_branch_mismatch tagged with namespace and reason and write a History-service warning log reading "Pagination branch token is not the execution's current branch token" with the fields wf-namespace, wf-id, wf-run-id, reason, wf-branch-token, and wf-request-branch-token. Search History warning logs for that message, or for reason=foreign_branch, which denotes a branch token the requested execution never recorded and is not expected in normal operation; reason=non_current_branch denotes a retained but superseded branch of the same execution and can occur legitimately after conflict resolution or failover. A foreign_branch record does not by itself prove the read was refused, because the counter and log are emitted before the enforcement decision and therefore appear both when the check rejects the request and when history.enablePaginationTokenBranchValidationShadowMode is enabled and the read is served; confirm refusal from the gRPC response, which is InvalidArgument carrying a CurrentBranchChangedFailure detail. The metric and log are tagged with the requesting namespace, not the namespace whose history the token selects, so a foreign_branch alert does not name the exposed namespace. The issue does not provide unauthenticated access, construction of a usable continuation from public workflow identifiers, enumeration of workflows for which the caller holds no continuation, or any modification of workflow state.
-
CVE-2026-61708
SOCFortress CoPilot focuses on providing a single pane of glass for all your security operations needs. Prior to 0.1.76, `POST /api/agent_data_store/upload` in `backend/app/data_store/data_store_routes.py` permits `customer_user` callers to choose `bucket_name` and `object_name` without an allowlist, tenant prefix, ownership check or content restriction. The only file check requires a nonempty filename. The operations in `backend/app/data_store/data_store_operations.py` create missing buckets and upsert objects, allowing existing objects belonging to any tenant to be overwritten. Targets include `sysmon-configs/<customer_code>/sysmon_config.xml` distributed to managed Windows endpoints, `copilot-case-report-templates/<name>` rendered server-side, and case evidence or forensic artifacts under `copilot-cases/<case_id>/<file>` and `velociraptor-artifacts/<agent_id>/<flow_id>/<file>`. A lowest-privilege user can tamper with cross-tenant evidence, blind endpoint telemetry, poison report templates for later code execution, plant data or exhaust storage. Sibling list, download and delete routes remain restricted to admin/analyst; the exposed operation is the upload write path. This issue is fixed in version 0.1.76.
-
CVE-2026-57207
NetBox MCP Server is a simple read-only Model Context Protocol server for NetBox. Prior to version 1.2.1, when running with TRANSPORT=http, the MCP server has no authentication on the HTTP endpoint. Any client that can reach the port gets a valid MCP session and full access to all four NetBox tools - no token or credential needed. This issue has been patched in version 1.2.1.
-
CVE-2026-127091
Jadaptive Maverick Synergy Server before 3.0.33 and 3.1.x before 3.1.7, and Maverick Legacy Server before 1.8.1, contain an improper path equivalence flaw in VirtualFileFactory.canonicalisePath that bypasses mount-level access policy. Authenticated users can supply paths with duplicate separators or dot segments to route operations to a less restrictive parent mount and read or write files the child mount denies.
-
CVE-2026-120818
A use-after-free weakness in MongoDB Server's server-side JavaScript support for aggregation may allow a stale reference to previously freed process memory to be read and returned. An authenticated user with minimal or no assigned roles can trigger this condition over the normal network protocol by submitting a single crafted aggregation request, with no additional privileges or non-default configuration. Successful triggering this issue may result in adjacent process memory contents being returned, including data that undermines memory-layout randomization, and may in some cases affect server availability.
-
CVE-2026-120010
SAP NetWeaver Application Server (SAP Message Server) does not properly size memory allocations when handling certain HTTP requests. An unauthenticated attacker could send a specially crafted request to trigger a heap buffer overflow in affected functionality. Successful exploitation causes the component to abort and remain unavailable until it is restarted by the framework, resulting in a high impact on availability of the system. There is no impact on confidentiality and integrity of the application.
-
CVE-2026-119078
Apache Commons Codec's Blake3 class documented the output of its doFinalize methods as depending on any previously finalized bytes, so that consecutive calls would return different output. In fact, doFinalize does not change the hash state: calling it again without supplying more input returns the same bytes. The key-derivation example in the class Javadoc, shipped since 1.16.0, derives two keys with two consecutive doFinalize calls. Code that follows it gets identical keys, so the separation between the derived keys is lost. The same applies to any application that calls doFinalize more than once on one instance and expects new output each time. This issue affects Apache Commons Codec from 1.16.0 before 1.23.0. Version 1.23.0 corrects the documentation, but the behavior of doFinalize is unchanged. Upgrading alone does not fix affected code. Users should check their applications for repeated doFinalize calls on the same Blake3 instance and derive multiple keys by requesting the combined length in one call and splitting the output, as the updated Javadoc shows.
-
CVE-2026-55614
uhttpd is a small HTTP server. In unpatched repository snapshots, `client_parse_header()` in `client.c` lowercases header names but compares the `Transfer-Encoding` value to `chunked` using case-sensitive `strcmp`. A mixed-case transfer-coding token is accepted without enabling `r->transfer_chunked`, despite transfer-coding tokens being case-insensitive. On a keep-alive connection, the `chunked` POST body is left unread and request-shaped body bytes are parsed as another HTTP request, permitting HTTP request smuggling and boundary desynchronization. Commits ae015e099986ceace44975cb69629841a2d58d37 and 6641b050e5f7c7fd9ae92e8e119d6d6dfc6e9837 patch the issue.
-
CVE-2026-19128
IBM API Connect 12.1.0.0 through 12.1.1.2 could allow an authenticated attacker to send unauthorized requests from the system due to server-side request forgery. This could be used to enumerate network resources or to support further attacks.
-
CVE-2026-93327
A Deserialization of Untrusted Data vulnerability in the SubWorkflow task in Google Cloud Application Integration versions prior to 2026-09-11 on Google Cloud Platform allows an authenticated attacker to execute arbitrary code using a maliciously crafted serialized object. This vulnerability was patched on 11 September 2026, and no customer action is needed.
-
CVE-2026-117565
Artalk before 2.10.1 contains an unauthenticated server-side request forgery vulnerability that allows remote attackers to make the server fetch arbitrary URLs by submitting anonymous comments to /api/v2/comments. Attackers can set a page_key to an internal URL, which GetTitleByURL fetches after only scheme validation, then read the page title through the public comment list.
-
CVE-2026-74756
Some docker-compose shipped by the project exposes by default Webadmin REST API publically without proper authentication setup. Version 3.9.1 onward sets a random credential by default on unprotected webadmin endpoints to prevent this while retain convenience.
-
CVE-2026-81938
Grafana Mimir's Alertmanager rejects tenant-supplied receiver configurations that reference local files, but the check did not cover two HTTP client settings: http_headers.<name>.files and basic_auth.username_file. A tenant permitted to upload an Alertmanager configuration could point these fields at any file readable by the Alertmanager process. When a notification was sent, Alertmanager read the file and put its contents in a header of the outgoing request. By directing the receiver at a server they control, the tenant could read files from the Alertmanager host, such as credentials and configuration. Exploitation requires permission to write the tenant's Alertmanager configuration. Grafana Enterprise Metrics (GEM), which is built on Mimir, is also affected. The issue is fixed in Mimir 3.1.7 and 3.2.2 and in GEM 2.17.20. GEM 2.11 through 2.16 will not receive a fix; users on those versions should upgrade to 2.17.20 or later.
-
CVE-2026-116191
Rclone before 1.75.2 contains a path traversal vulnerability in serve webdav that allows unauthenticated attackers to write above the served root by sending an unvalidated '..' path component. Attackers can send MKCOL /.. to plant a VFS node, then PUT, MOVE, COPY, or MKCOL to create, overwrite, or delete files belonging to other tenants.
-
CVE-2026-115788
IBM App Connect Operator and IBM App Connect Enterprise Certified Containers Operands is vulnerable to Log Forging on a request to create a resource.
-
CVE-2026-121885
A flaw was found in Podman. An attacker can trigger a Denial of Service (DoS) by supplying a container image containing crafted metadata labels. Podman's auto-update mechanism improperly trusts these image labels to determine which system services to restart. If a user runs the malicious container and auto-update processes it, arbitrary system services are restarted with the user's privileges, potentially allowing an attacker to shut down the host system when Podman runs with administrative privileges.
-
CVE-2026-113504
Apache Airflow's secrets masker stopped descending into nested structures after five levels, so a secret registered with the masker — for example a Connection password or a Variable value — that appeared more than five containers deep under non-sensitive keys was shown in cleartext in task logs and rendered template fields. A user who can read task logs or rendered templates, but is not meant to see those secrets, could read them. Users are advised to upgrade to apache-airflow 3.4.0 or later. Airflow 2.x is also affected; Airflow 2 has reached end of life and will not receive a fix, so Airflow 2 users should upgrade to Apache Airflow 3.4.0 or later.
-
CVE-2026-97677
IBM Langflow OSS 1.0.0 through 1.12.2 An authenticated flow author could write attacker-controlled file content into any directory writable by the service account, bypassing the platform's local file access isolation safeguard. Via a crafted on-disk index, the attacker could also read the contents of arbitrary files accessible to the service process — including configuration files, secrets, or database files stored under the application's data directory.
-
CVE-2026-77604
Code Graph RAG is an AI-powered knowledge graph and retrieval system for querying, understanding, and editing multi-language codebases. Prior to 0.0.345, the MCPToolsRegistry.read_file implementation in codebase_rag/mcp/tools.py joins attacker-controlled file_path input to project_root without resolving and verifying containment when offset or limit selects the paginated branch. An absolute path causes pathlib to discard the configured root, while relative traversal can also resolve outside it. The protected non-paginated branch uses validate_project_path, but the paginated branch bypasses that validation before opening the path. An unauthenticated caller of a network-exposed HTTP MCP server can therefore read arbitrary UTF-8-readable files available to the server process, including credentials, environment data, private keys, source, and configuration; local callers through the stdio transport can reach the same vulnerable code path. This issue is fixed in version 0.0.345.
-
CVE-2026-59179
OpenHop is a tool for creating interactive data-flow diagrams. Prior to 0.3.7, `@openhop/server` passes unauthenticated flow route parameters from `GET /api/flows/:id` and `DELETE /api/flows/:id` in `packages/server/src/routes.ts` to `FlowStore.filePath()` in `packages/server/src/store.ts` without an identifier allow-list. The router applies `decodeURIComponent` to the id, and `path.join()` then normalizes traversal outside the configured flow directory after appending the .yaml extension. An attacker can read accessible YAML files as YAML-parsed JSON or permanently delete accessible .yaml files outside that directory, exposing configuration or secrets and disrupting dependent services. In `packages/server/src/index.ts`, CORS `origin: true` permits any browser origin, enabling a malicious webpage to target a loopback-bound instance. Docker deployments default to `HOST=0.0.0.0` and permit direct unauthenticated network access. This issue is fixed in repository release 0.3.7.
-
CVE-2026-114928
Vulnerability in the PeopleSoft Enterprise PeopleTools product of Oracle PeopleSoft (component: Application Server). Supported versions that are affected are 8.61-8.63. Easily exploitable vulnerability allows low privileged attacker with network access via HTTP to compromise PeopleSoft Enterprise PeopleTools. Successful attacks of this vulnerability can result in takeover of PeopleSoft Enterprise PeopleTools. CVSS 3.1 Base Score 8.8 (Confidentiality, Integrity and Availability impacts). CVSS Vector: (CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H).
-
CVE-2026-125275
Loop with Unreachable Exit Condition ('Infinite Loop') vulnerability in device certificate verification in nerves-hub nerves_hub_web allows an unauthenticated remote attacker to cause a denial of service via a TLS client certificate whose CA is not registered. When a client certificate chains to a root in the device endpoint's trust store but no CA in the chain is registered in NervesHub, NervesHub.SSL.decide/2 refuses it with the reason :unknown_ca. Erlang/OTP ssl reads that reason as a missing trust anchor, looks for an alternative root, finds the same root and validates again without end. The handshake process queries the database on each pass and keeps running after the client disconnects, until the node restarts. The device endpoint trusts the public roots from CAStore unless DEVICE_SSL_CACERTFILE is set, so a client certificate from a public CA whose chain allows client authentication is sufficient. A chain that ends at a root outside the trust store is refused once. This issue affects nerves_hub: from 1.0.0 before 2.5.4.
-
CVE-2026-107518
Dell PowerStore contains a Server-Side Request Forgery (SSRF) vulnerability. An authenticated user with limited privileges could potentially exploit this vulnerability to make the appliance connect to attacker-chosen internal systems.
-
CVE-2026-118152
Cross-site request forgery (csrf) in Microsoft Exchange Server allows an unauthorized attacker to perform spoofing over a network.
-
CVE-2026-108946
Improper input validation vulnerability in Apache ActiveMQ, Apache ActiveMQ KahaDB. The ActiveMQ broker resolves connection, session and consumer state from ids that are user provided and in some scenarios does not validate that they ids are bound to that user. This allows an authenticated client to spoof another user's ids. This allows: * an authenticated client consumes from another connection's temporary queue/topic by putting the victim's connection id in its own ConsumerId, bypassing the broker-side ownership check. * a duplicate ConsumerId registration is only logged, after which the attacker's RemoveInfo or disconnect tears down the victim's live subscription * a forged MessageAck for the victim's ConsumerId permanently deletes an in-flight, unacknowledged message. This issue affects Apache ActiveMQ: from 6.0.0 before 6.3.3, before 5.19.12; Apache ActiveMQ KahaDB: from 6.0.0 before 6.3.3, before 5.19.12. Users are recommended to upgrade to version 6.4.0, 6.3.3, or 5.19.12, which fixes the issue.
-
CVE-2026-119049
FreeMail before 5.3.2 contains an authorization bypass vulnerability that allows authenticated users to claim another user's existing mailbox via the POST /api/create endpoint. Attackers who know a victim's address can become co-owners through assignMailboxToUser to read messages and verification codes, set forwarding, and delete mail.
-
CVE-2026-115646
Partial String Comparison Vulnerability in HiRDB. This issue affects HiRDB Server Version 10: from 10-00 before 10-12-01; HiRDB Server Version 9: from 09-50-06 through 09-66-22; HiRDB Server with Additional Function Version 9: from 09-50-/F through 09-66/V; HiRDB Server Version 9(32): from 09-50-06 through 09-66-22; HiRDB Server with Additional Function Version 9(32): from 09-50-/F through 09-66-/V.
-
CVE-2026-61789
DBHub is a database MCP server for Postgres, MySQL, SQL Server, Oracle, MariaDB, SQLite. Prior to 0.22.6, read-only SQL validation can be bypassed on the MySQL and MariaDB connectors. In `src/utils/sql-parser.ts`, `scanSingleLineComment` treats every double dash as a comment start, whereas MySQL and MariaDB require following whitespace, a control character, or the end of the line. The read-only check in `src/tools/execute-sql.ts` validates split statements with `isReadOnlySQL`, but `src/connectors/mysql/index.ts` and `src/connectors/mariadb/index.ts` execute the original string with `multipleStatements` enabled. A double dash followed by a non-whitespace character can therefore hide a subsequent write or destructive DDL statement from the validator while the database interprets the dashes as minus operators and executes both statements. Anyone able to influence the assistant's SQL, including through prompt injection in content the assistant reads, can bypass read-only mode to modify or drop database data, subject to the database account's privileges. PostgreSQL is not affected by this path because its double-dash parsing agrees with the checker and its connector executes individual split statements. This issue is fixed in version 0.22.6.
-
CVE-2026-103963
A flaw was found in glib-networking. When establishing connections using Transport Layer Security (TLS) version 1.3, the client-side session cache does not limit the number of session tickets retained for a single host. A malicious server can exploit this by sending excessive session ticket messages, causing the client application to continuously consume memory. This unbounded memory growth can exhaust system memory, resulting in a Denial of Service (DoS).
-
CVE-2026-115810
Tandoor Recipes before 2.7.0 contains a server-side request forgery vulnerability in the Pestle importer that allows authenticated low-privileged users to force the server to issue GET requests to arbitrary internal or external hosts by crafting a malicious image URL in a Pestle JSON import file. Attackers can exploit the raw requests.get call in cookbook/integration/pestle.py, which bypasses the application's centralized requests-hardened SSRF filter, to reach internal-only endpoints, perform port/service discovery, and exfiltrate internal responses via the stored recipe image.
-
CVE-2026-109158
IBM App Connect Operator and IBM App Connect Enterprise Certified Containers Operands may allow the AI Agent to log sensitive information about the deployed environment within error messages.
-
CVE-2026-110314
Client-Side Enforcement of Server-Side Security vulnerability in the WebSocket connection handling of mtrudel bandit allows a remote client to keep running WebSock handler code after the server has stopped its connection, via data frames sent after ignoring the server's Close frame. When a WebSock callback returns {:stop, ...}, Bandit.WebSocket.Connection calls terminate/2, sends Close and shuts down the write side, but keeps the connection process running in a :closing state. handle_frame/3 and handle_info/3 never check that state, so later frames still reach handle_in/2 and later messages reach handle_info/2, with the state from the stop tuple, until the client sends Close, TCP drops or the idle timeout fires. A client refused in init/1 after a failed token check could still run an admin-only action. This issue affects bandit: from 0.5.1 before 1.14.0.
-
CVE-2026-91469
Vulnerability in the Oracle WebLogic Server product of Oracle Fusion Middleware (component: Core). Supported versions that are affected are 12.2.1.4.0, 14.1.1.0.0, 14.1.2.0.0 and 15.1.1.0.0. Easily exploitable vulnerability allows unauthenticated attacker with network access via HTTP to compromise Oracle WebLogic Server. Successful attacks of this vulnerability can result in unauthorized creation, deletion or modification access to critical data or all Oracle WebLogic Server accessible data as well as unauthorized read access to a subset of Oracle WebLogic Server accessible data and unauthorized ability to cause a hang or frequently repeatable crash (complete DOS) of Oracle WebLogic Server. CVSS 3.1 Base Score 9.4 (Confidentiality, Integrity and Availability impacts). CVSS Vector: (CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:L/I:H/A:H).
-
CVE-2026-123487
Parse Server before 8.6.115 and 9.0.0 before 9.10.7-alpha.1 validates user email address format with a regular expression whose run time grows quadratically with input length and that is applied without a length limit. An unauthenticated attacker who knows the application ID can send a sign-up request, or a logged-in user can update their own email, with a long crafted email address to block the Node.js event loop and deny service to all clients.