Security vulnerabilities, privacy incidents, safety concerns, and policy updates affecting LLMs and AI agents.
Langflow's /profile_pictures/{folder_name}/{file_name} endpoint has a path traversal vulnerability (a flaw where attackers use ../ sequences to access files outside the intended directory). The folder_name and file_name parameters aren't properly validated, allowing attackers to read the secret_key file across directories. Since the secret_key is used for JWT authentication (a token system that verifies who you are), an attacker can forge login tokens and gain unauthorized access to the system.
The h3 library's EventStream class fails to remove carriage return characters (`\r`, a line break in the Server-Sent Events protocol) from `data` and `comment` fields, allowing attackers to inject fake events or split a single message into multiple events that browsers parse separately. This bypasses a previous fix that only removed newline characters (`\n`).
etcd (a distributed key-value store used in systems like Kubernetes) has multiple authorization bypass vulnerabilities that let unauthorized users call sensitive functions like MemberList, Alarm, Lease APIs, and compaction when the gRPC API (a communication protocol for remote procedure calls) is exposed to untrusted clients. These vulnerabilities are patched in etcd versions 3.6.9, 3.5.28, and 3.4.42, and typical Kubernetes deployments are not affected because Kubernetes handles authentication separately.
Langflow has a vulnerability where the image download endpoint (`/api/v1/files/images/{flow_id}/{file_name}`) allows anyone to download images without logging in or proving they own the image (an IDOR, or insecure direct object reference, where attackers access resources by manipulating identifiers). An attacker who knows a flow ID and filename can retrieve private images from any user, potentially exposing sensitive data in multi-tenant setups (systems serving multiple separate customers).
PinchTab is an HTTP server (a program that handles web requests) that lets AI agents control a Chrome web browser. Versions 0.8.2 and earlier have a blind SSRF vulnerability (a flaw where an attacker tricks the server into making requests to internal networks that should be off-limits) in the /download endpoint, because the server only checks the URL once but the browser can follow hidden redirects to reach internal addresses. The risk is limited because the vulnerable feature is disabled by default.
FastGPT (an AI platform for building AI agents) versions 4.14.8.3 and below have a critical security flaw where the fastgpt-preview-image.yml workflow uses pull_request_target (a GitHub feature that runs code with access to repository secrets) but executes code from an external contributor's fork, allowing attackers to run arbitrary code (commands on systems they don't own), steal secrets, and potentially compromise the production container registry (the central storage system for packaged software).
SQLBot, an intelligent data query system that uses a large language model and RAG (retrieval-augmented generation, where an AI pulls in external documents to answer questions), has a critical SQL injection vulnerability (a bug where an attacker tricks the system into running unintended database commands) in versions before 1.7.0 that allows authenticated users to execute arbitrary code on the backend server. The vulnerability exists because Excel sheet names are directly inserted into database commands without proper sanitization (cleaning/validation), and attackers can exploit this by uploading specially crafted files to gain complete control of the system.
SQLBot, an AI-based system for querying databases that uses RAG (retrieval-augmented generation, where it pulls in external data to answer questions), has a vulnerability in versions before 1.7.0 that lets attackers read any file from the server. An attacker can exploit the /api/v1/datasource/check endpoint by submitting a fake MySQL connection with a malicious setting, which tricks the server into reading and sending back sensitive files like /etc/passwd when it tries to verify the connection.
SQLBot, a data query system combining AI with RAG (retrieval-augmented generation, where an AI pulls in external documents to answer questions), has a critical vulnerability in versions 1.5.0 and below that chains three security gaps: missing permission checks on file uploads, unsanitized storage of user input, and inadequate protections when inserting data into the AI's instructions. An attacker can exploit this to trick the AI into running malicious database commands that give them control over the database server.
Discourse, an open-source discussion platform, has a cross-site scripting vulnerability (XSS, where attackers inject malicious code that runs in a user's browser) in versions before 2026.3.0-latest.1, 2026.2.1, and 2026.1.2. The vulnerability exists because the system trusts output directly from an AI language model and displays it without proper sanitization (cleaning) in the Review Queue interface, allowing attackers to use prompt injection (tricking the AI by hiding instructions in user input) to make the AI generate malicious code that executes when staff members review flagged posts.
CVE-2026-26137 is a server-side request forgery vulnerability (SSRF, a flaw where an attacker tricks a server into making unwanted network requests on their behalf) in Microsoft 365 Copilot's Business Chat that allows an authorized attacker to gain elevated privileges over a network. The vulnerability affects an exclusively hosted service and was published on March 19, 2026.
CVE-2026-26136 is a command injection vulnerability (a flaw where an attacker can insert malicious commands by exploiting improper filtering of special characters) in Microsoft Copilot that allows an unauthorized attacker to access and disclose sensitive information over a network.
CVE-2026-24299 is a command injection vulnerability (a flaw where an attacker can insert malicious commands into an application by exploiting improper handling of special characters) in Microsoft 365 Copilot that allows an unauthorized attacker to disclose information over a network. The vulnerability has a CVSS 4.0 severity rating (a 0-10 scale measuring how serious a security flaw is). This is hosted exclusively as a service by Microsoft.
Langflow's file upload endpoint (POST /api/v2/files/) is vulnerable to arbitrary file write (a type of attack that lets attackers save files anywhere on a server) because it doesn't properly validate filenames from multipart requests. Attackers who are logged in can use directory traversal characters (like "../") in filenames to write files outside the intended directory, potentially achieving RCE (remote code execution, where attackers can run commands on the server).
Claude Code had a security flaw where it would read settings from a file (`.claude/settings.json`) that could be controlled by someone creating a malicious repository, allowing them to bypass the workspace trust dialog (a security prompt that asks for permission before running code). This meant an attacker could trick users into running code without their knowledge or consent. The vulnerability has been patched.
MLflow, a machine learning platform, has a vulnerability (CVE-2025-15031) in how it extracts model files from compressed archives. The issue is that the software uses `tarfile.extractall` (a Python function that unpacks compressed tar files) without checking whether file paths are safe, allowing attackers to use specially crafted archives with `..` (parent directory references) or absolute paths to write files outside the intended folder. This could let attackers overwrite files or execute malicious code, especially in shared environments or when processing untrusted model files.
Mesop contains a critical vulnerability in its testing module where a `/exec-py` route accepts Python code without any authentication checks and executes it directly on the server. This allows anyone who can send an HTTP request to the endpoint to run arbitrary commands on the machine hosting the application, a flaw known as unauthenticated remote code execution (RCE, where an attacker runs commands on a system they don't own).
Fix: Upgrade to etcd 3.6.9, etcd 3.5.28, or etcd 3.4.42. If upgrading is not immediately possible, restrict network access to etcd server ports so only trusted components can connect, and require strong client identity at the transport layer such as mTLS (mutual TLS, where both client and server verify each other's identity) with tightly scoped client certificate distribution.
GitHub Advisory DatabaseFix: The issue has been patched in version 0.8.3.
NVD/CVE DatabaseFix: Update to version 1.7.0 or later, where this issue has been fixed.
NVD/CVE DatabaseFix: Update to version 1.7.0 or later. The source states: 'This issue was fixed in version 1.7.0.'
NVD/CVE DatabaseLaravel Livewire (a PHP framework for building interactive web applications) contains a code injection vulnerability (a flaw where attackers can insert malicious code into an application) that allows unauthenticated attackers to execute arbitrary commands on affected systems in certain situations. This vulnerability is currently being actively exploited by attackers in the wild.
Fix: Apply mitigations per vendor instructions, follow applicable BOD 22-01 guidance for cloud services, or discontinue use of the product if mitigations are unavailable. The due date for remediation is 2026-04-03.
CISA Known Exploited VulnerabilitiesApple's operating systems (watchOS, iOS, iPadOS, macOS, visionOS, and tvOS) contain an improper locking vulnerability (a flaw that fails to properly control access to shared memory between processes), which allows a malicious application to make unexpected changes to memory that multiple programs use. This vulnerability is currently being exploited by attackers in real-world attacks.
Fix: Apply mitigations per Apple's vendor instructions using the provided support links, follow applicable BOD 22-01 guidance for cloud services, or discontinue use of the product if mitigations are unavailable. The due date for remediation is 2026-04-03.
CISA Known Exploited VulnerabilitiesA buffer overflow vulnerability (a programming error where data overflows its allocated memory space) affects multiple Apple products including watchOS, iOS, iPadOS, macOS, visionOS, and tvOS. A malicious app could exploit this to crash the system or write malicious code directly into kernel memory (the core of the operating system). This vulnerability is actively being exploited by attackers in the wild.
Fix: Apply mitigations per Apple's vendor instructions (referenced in support documents), follow applicable BOD 22-01 guidance for cloud services, or discontinue use of the product if mitigations are unavailable. The deadline for remediation is April 3, 2026.
CISA Known Exploited VulnerabilitiesFix: The issue is fixed in v1.6.0.
NVD/CVE DatabaseFix: Update to versions 2026.3.0-latest.1, 2026.2.1, or 2026.1.2, which contain a patch. Alternatively, as a workaround, temporarily disable AI triage automation scripts.
NVD/CVE DatabaseFix: The source recommends two fixes: (1) Sanitize the multipart filename by extracting only the file name component and rejecting names containing "..": `new_filename = StdPath(file.filename or "").name` and add validation to reject invalid names. (2) Add a canonical path containment check inside `LocalStorageService.save_file` using `resolve().is_relative_to(base_dir)` to ensure files are always saved within the intended base directory.
GitHub Advisory DatabaseFix: Users on standard Claude Code auto-update have already received the fix. Users performing manual updates are advised to update to the latest version.
GitHub Advisory Database