Intelligence
Explorer.

Tesseract is an open source OCR engine. In version 5.5.3 and earlier, RecodedCharID::DeSerialize in src/ccutil/unicharcompress.h validates length_ but accepts negative code_ values from a crafted .traineddata recoder component. UnicharCompress::ComputeCodeRange in src/ccutil/unicharcompress.cpp can consequently produce code_range_ equal to zero, after which SetupDecoder indexes is_valid_start_ with the negative code on a size-zero vector. The resulting out-of-bounds bit write uses a large wrapped index and reliably causes a wild-address crash or allocation failure on the default LSTM engine. No fixed release is available as of this review.

Tesseract is an open source OCR engine. In version 5.5.3 and earlier, prior .traineddata hardening added bounds checks to NetworkIO::CopyTimeStepGeneral and NetworkIO::Randomize in src/lstm/networkio.cpp but left NetworkIO::WriteTimeStepPart and NetworkIO::AddTimeStepPart unchecked. In LSTM::Forward in src/lstm/lstm.cpp, source_ is sized from the independently deserialized na_ field while the WriteTimeStepPart count is ns_, which comes from the CI gate WeightMatrix dim1() value. A crafted NT_LSTM layer can make ns_ much larger than na_, causing a heap out-of-bounds write during the first recognition step on the default LSTM engine and resulting in heap corruption, a crash, or potentially controlled corruption. No fixed release is available as of this review.

Tesseract is an open source OCR engine. In version 5.5.3 and earlier, FullyConnected::DeSerialize in src/lstm/fullyconnected.cpp does not validate the deserialized layer scalars ni_ and no_ against the weight-matrix dimensions. During FullyConnected::Forward, MatrixDotVector in src/lstm/weightmatrix.cpp writes w.dim1() results into temp_line, which is sized from no_, and reads w.dim2() minus one inputs from curr_input, which is sized from ni_. A crafted .traineddata NT_SOFTMAX layer can therefore use inconsistent dimensions to cause a heap out-of-bounds write and read on the default LSTM engine, resulting in heap corruption, a crash, information disclosure, or potentially controlled corruption. No fixed release is available as of this review.

Tesseract is an open source OCR engine. In version 5.5.3 and earlier, Classify::ReadNormProtos in src/classify/normmatch.cpp parses the NORMPROTO component of a .traineddata file and uses std::istream::operator>>(char*) to extract a whitespace-delimited token into a fixed 61-byte stack buffer without setting a stream width. The 100-byte line buffer can carry a token of up to 99 characters, so a token longer than 60 characters writes up to 39 attacker-controlled bytes past the buffer during TessBaseAPI::Init of the legacy engine, causing stack corruption, denial of service, and potentially control-flow hijacking on affected standard-library implementations. Builds using Apple's libc++ C++20 bounded array overload are incidentally protected, while typical libstdc++ builds remain affected. No fixed release is available as of this review.

rclone is a command-line program to sync files and directories to and from different cloud storage providers. Prior to 1.75.1, rclone core does not reject parent-directory segments in source Object.Remote() values before fs/list, fs/walk, fs/sync, and fs/operations pass those values to destination backends. A flat-keyspace source object store populated with native non-rclone tooling can contain a raw .. key segment, and affected b2, swift, qingstor, oracleobjectstorage, internetarchive, smb, storj, sftp, webdav, ftp, filelu, shade, and sia destinations use path.Join(root, remote) before EncodeDot can neutralize the segment. A copy or upload can therefore escape the configured root into another bucket, share, or path reachable by the victim credential, with sftp and smb potentially reaching other filesystem or share locations under the same login authority. This issue is fixed in version 1.75.1.

rclone is a command-line program to sync files and directories to and from different cloud storage providers. From 1.75.0 until 1.75.1, the serve S3 streamed multipart path in cmd/serve/s3/multipart.go passes attacker-controlled contentLength to multipart.NewRW().Reserve before reading request-body bytes. waitForTurn admits the current part and one oversized part when the buffer is empty despite --multipart-streaming-buffer-limit, and lib/pool allocates 1 MiB pages according to Content-Length or X-Amz-Decoded-Content-Length. A network client can retain or multiply these reservations without sending the declared body, exhausting process or host memory or permanently blocking request handlers. Anonymous S3 deployments require no credentials, while deployments using auth_key require an accepted S3 key. This issue is fixed in version 1.75.1.

rclone is a command-line program to sync files and directories to and from different cloud storage providers. From 1.70.0 until 1.75.1, the serve/start RC interface accepts per-server proxyOpt.AuthProxy settings, and the FTP and S3 constructors in cmd/serve/ftp/ftp.go and cmd/serve/s3/server.go incorrectly check the process-global proxy.Opt.AuthProxy value instead. When the global value is empty, the request-local authentication proxy is ignored: FTP falls back to the fixed filesystem with username anonymous and any password, while S3 with AuthKey serves the fixed RC fs rather than the backend selected by the proxy. The dedicated command-line servers that configure the global option are not affected. This issue is fixed in version 1.75.1.

An integer overflow in the tensor buffer validation component in Amazon Deep Java Library (DJL) from 0.13.0 through 0.36.0 on all platforms might allow a remote unauthenticated actor to obtain information from adjacent process memory or cause a denial of service via a crafted tensor payload. To remediate this issue, users should upgrade to version 0.37.0 or above.

FileRun before 2026.3.0 contains a PHP object injection vulnerability that allows authenticated attackers to execute arbitrary code by exploiting incorrect options passed to unserialize() in the Perms::getPerms() method, where a positional array is used instead of the required named-key array to disable class instantiation. Attackers with database write access can inject a serialized gadget chain into the permissions table columns processed on every authenticated page load to write arbitrary files, such as PHP webshells, to web-accessible paths.

FileRun before 2026.3.0 contains a SQL injection vulnerability that allows delegated or simple administrators to execute arbitrary SQL by submitting the description parameter as an array, causing the getValuesString() method in DB/DP.php to interpolate raw array values directly into an INSERT statement without parameterization. Because the underlying PDO connection uses emulated prepared statements enabling stacked queries, attackers can manipulate the df_users_permissions table to escalate a delegated administrator account to superuser privileges, and may additionally achieve code execution via unsanitized path values passed to require_once in the logs listing component.

FileRun before 2026.3.0 contains an OS command injection vulnerability caused by a no-op redefinition of escapeshellcmd() in CLI.php that strips shell-metacharacter escaping, allowing attacker-controlled input to reach an exec() sink unsanitized. Attackers can exploit this through an interactive path via image_preview.php with a crafted args parameter requiring superuser authentication, or through a persistent path by storing malicious payloads in thumbnails_ffmpeg_args or thumbnails_ffmpeg_ss that execute when any user triggers video thumbnail generation.

FileRun before 2026.3.0 contains an OS command injection vulnerability in the PhotoProofSheet handler that allows authenticated users with upload permission to execute arbitrary commands by uploading files with shell metacharacters in their names. Attackers can upload a file containing command substitution syntax such as backticks, semicolons, or $() sequences in the filename, then trigger the PhotoProofSheet endpoint to execute arbitrary commands as the web-server user due to missing escapeshellarg() sanitization in the ImageMagick montage command construction.

A Time-of-check Time-of-use (TOCTOU) race condition leading to insecure symlink following in Plesk causes local privilege escalation to root via arbitrary file/directory ownership takeover.

Path traversal in Plesk's Backup Manager causes arbitrary file write as root by an authenticated customer.

OS command injection in the advanced-rule parser of ConfigServer Security & Firewall allows a remote attacker who controls a configured allow/deny feed to execute arbitrary commands as root, due to insufficient validation of feed-supplied rule data. The vulnerability affects versions of the software originally distributed by ConfigServer, as well as versions of the WebPros-maintained fork that contain the vulnerable code. WebPros has addressed the vulnerability in version 16.30. Other forks or independently maintained versions of ConfigServer Security & Firewall (CSF) may also be affected and should be evaluated independently.

Improper escaping of a request URL in ConfigServer Security & Firewall allows an unauthenticated remote attacker to execute arbitrary commands as the CSF service account via shell command injection. The vulnerability affects versions of the software originally distributed by ConfigServer, as well as versions of the WebPros-maintained fork that contain the vulnerable code. WebPros has addressed the vulnerability in version 16.30. Other forks or independently maintained versions of ConfigServer Security & Firewall (CSF) may also be affected and should be evaluated independently.

An issue in Flowise 3.1.2 allows a remote attacker to execute arbitrary code via the /api/v1/prediction/<flowId> endpoint

An issue in AppFlowy 0.11.8 allows a remote attacker to execute arbitrary code via the afLaunchUri, _afLaunchLocalUri (url_launcher.dart), OpenFilex.open, localPathRegex (common_patterns.dart) components

Anchor CMS through 0.12.7 fails to enforce role-based access control in admin user-management endpoints, allowing any authenticated low-privilege user to create administrator accounts or modify existing ones. Attackers with editor or user roles can POST directly to admin/users/add or admin/users/edit endpoints to create new administrator accounts or change the existing administrator's password, gaining full administrative access.

knowns through 0.33.0 fails to validate the path query parameter in the workspace browse endpoint, allowing remote attackers to enumerate arbitrary directories on the host filesystem. Attackers can traverse the directory structure to locate project directories and identify targets for further exploitation.

knowns through 0.33.0 exempts the project.set action from permission guard checks unconditionally, allowing read-only agent sessions to bypass restrictions. Attackers can invoke project.set to repoint the server at another project directory and obtain write access capabilities.

knowns through 0.33.0 fails to confine the path argument of the code.find MCP tool to the project root, allowing AI agent sessions to read source files anywhere on the host. Attackers can supply absolute paths or relative traversal sequences to the path argument and retrieve full file contents from outside the intended project directory.

knowns through 0.33.0 fails to properly validate template destination paths in the code generation template engine, allowing attackers to read and write arbitrary files outside the project root. Attackers can supply malicious templates that traverse directories to overwrite shell profiles, steal credentials, or achieve persistent code execution on victim systems.

knowns versions before 0.31.0 fail to properly validate the x-opencode-directory request header in the /api/opencode proxy endpoint. Remote attackers can supply arbitrary directory paths to execute file operations outside the project root on the host system.