By weakness (CWE)

CWE-693: related vulnerabilities

CVEs classified under CWE-693. Understanding the weakness class helps prioritize systemic fixes over one-off patches.

88 published vulnerabilities

• CVE-2025-24284HIGH 8.8

  CVE-2025-24284 is a sandbox escape vulnerability in macOS that allows a malicious or compromised application to break out of Apple's application sandbox and gain unauthorized access to system resources. An attacker with local access—or through a compromised app—could circumvent Apple's security boundaries to execute privileged actions outside the intended application container. The flaw was resolved in macOS Sequoia 15.4 through improved authorization checks.

• CVE-2025-71322HIGH 8.8

  PickleScan is a security tool designed to inspect Python pickle files and block dangerous code execution. Before version 0.0.33, it had a critical oversight: it failed to block the pty.spawn function, which allows spawning new processes. An attacker can exploit this gap by crafting a malicious pickle file that uses pty.spawn to run arbitrary commands when PickleScan processes it. The vulnerability is particularly dangerous because it requires minimal interaction—just opening a file—and can be delivered remotely.

• CVE-2026-11248HIGH 8.8

  CVE-2026-11248 is a bypass vulnerability in Google Lens, a feature within Chrome that allows users to perform visual searches. An attacker can craft a malicious webpage that, when visited by a user, circumvents Chrome's navigation security controls. This means a user could be redirected to an unintended destination or prevented from safely navigating away. The vulnerability requires user interaction—the user must visit the attacker's page—but once there, the attack happens automatically. Google has patched this in Chrome 149.0.7827.53 and later.

• CVE-2026-14535HIGH 8.8

  Trail of Bits fickling is a Python library designed to safely deserialize pickle files by analyzing them for dangerous imports before unpickling. A logic flaw in versions up to 0.1.11 breaks one of its two main safety checks. When fickling examines a pickle file, it runs two separate inspection passes: one flags obviously dangerous imports, and a second one (MLAllowlist) is supposed to catch sneaky imports from non-ML libraries that the first pass missed. However, the first pass leaves markers in shared memory that trick the second pass into skipping its checks entirely. As a result, any import from Python's standard library—except those on a small blocklist—gets a green light, even if it's dangerous. Because fickling's public API treats a green light as permission to actually deserialize and run the pickle, an attacker can craft a pickle file that imports and executes arbitrary standard library code, bypassing the security gate entirely.

• CVE-2026-57280HIGH 8.8

  Jenkins Script Security Plugin versions up to 1402.v94c9ce464861 contain a sandbox bypass vulnerability that allows authenticated attackers to execute arbitrary code. The flaw stems from incomplete type-cast handling in Groovy for-each loops, which permits attackers to invoke constructors that should be restricted by the sandbox. An attacker with script-execution privileges can exploit this to escape the plugin's security controls and run arbitrary Java constructors on the Jenkins server.

• CVE-2026-47135HIGH 8.7

  vm2 is a popular Node.js sandbox designed to safely execute untrusted code. A flaw in how it protects against dangerous JavaScript symbols allows attackers to bypass the sandbox and execute code with the privileges of the parent Node.js process. An attacker can craft malicious code that obtains cross-realm symbols (special JavaScript objects that exist in the host context), inject them into host objects, and hijack critical host-side functionality—such as the util.promisify method—to gain control over the host environment. The vulnerability affects all versions before 3.11.4.

• CVE-2026-47139HIGH 8.6

  vm2 is a popular Node.js sandbox designed to safely execute untrusted code in an isolated environment. A flaw in versions before 3.11.4 allows bypassing network restrictions by exploiting Node.js internal HTTP modules that the sandbox wasn't designed to block. Even when administrators explicitly deny access to standard network libraries like http and https, attackers can abuse hidden internal modules (_http_client and _http_server) to make outbound requests or create listening sockets, defeating the sandbox's security controls.

• CVE-2026-47209HIGH 8.6

  vm2 is a sandbox for Node.js that isolates untrusted code. A bug in how it handles property assignments on inherited objects allows attackers to bypass its security boundaries and write dangerous properties to the host system. Instead of properly creating properties on child objects, the sandbox writes directly to the host, letting malicious code inject capabilities that could compromise the Node.js runtime itself.

• CVE-2026-48721HIGH 8.6

  Warp, an AI-assisted development environment, contains a security flaw that allows attackers to bypass safety restrictions on dangerous commands. The vulnerability exists because Warp's non-interactive CLI agent checks whether commands are 'blocked' before properly handling environment variables at the start of command strings. An attacker who can control what the agent outputs—such as through a compromised repository or malicious input—can craft commands that evade the safety denylist. This could allow execution of commands that normally require user confirmation. The flaw affects versions from October 2025 through early May 2026 and is patched in the May 2026 release.

• CVE-2026-54762HIGH 8.6

  Traefik, a widely-used reverse proxy and load balancer, has a flaw in how it handles Kubernetes authentication configuration. When an operator sets up BasicAuth or DigestAuth protection on a route via Kubernetes annotations but the system cannot find or read the required authentication credential (the Secret), Traefik logs the error and then publishes the route anyway—without the authentication protection. This means unauthenticated users can access backends that operators believed were protected. The flaw affects versions 3.7.0-ea.1 through 3.7.4 and is corrected in 3.7.5.

• CVE-2026-12031HIGH 8.3

  Google Chrome on Windows contains a flaw in how it handles Views that could allow an attacker with a compromised renderer process to escape the browser sandbox. An attacker would need to first compromise Chrome's renderer—typically through a separate browser vulnerability—then serve a specially crafted HTML page to trigger the sandbox escape. This is a serious chaining vulnerability because successful exploitation grants access to the operating system with the privileges of the user running Chrome.

• CVE-2026-12438HIGH 8.3

  A flaw in how Google Chrome handles web content on Android devices could allow an attacker to escape the browser's security sandbox if they first compromise Chrome's rendering process. The attacker would need to craft a malicious HTML page to trigger the vulnerability. While Chrome's sandbox is designed to isolate web content from the underlying system, this weakness creates a path to break out of that isolation—a severe issue because it could lead to full device compromise. The vulnerability affects Chrome versions before 149.0.7827.155 on Android.

• CVE-2026-13951HIGH 8.3

  A vulnerability in Google Chrome's USB policy enforcement allows an attacker who has already compromised the browser's renderer process—the component that interprets web content—to potentially break out of Chrome's sandbox security boundary. The attacker would need to trick a user into visiting a maliciously crafted webpage. Once the renderer is compromised, this flaw could enable unauthorized access to the system, modification of data, or system unavailability. The issue affects Chrome versions before 150.0.7871.47.

• CVE-2026-14151HIGH 8.3

  Google Chrome versions prior to 150.0.7871.47 contain a flaw in how its AI components handle certain operations, which could allow an attacker to escape the browser's sandbox security boundary. The attack requires two conditions: first, the attacker must already have compromised Chrome's renderer process (the part that displays web pages), and second, the user must interact with a malicious web page. If both conditions are met, the attacker could gain unauthorized access to the system beyond what the sandbox normally permits. While Google rates this as low severity internally, the potential impact—full system compromise through sandbox escape—warrants urgent patching.

• CVE-2026-53853HIGH 8.3

  OpenClaw versions prior to 2026.5.12 contain a security flaw that allows authenticated users to bypass restrictions on which command-line arguments can be passed to approved executable programs. An attacker with login access can invoke whitelisted applications with arguments that should be forbidden, potentially gaining unauthorized access to files, establishing network connections, or executing additional commands that the OpenClaw security policy was designed to prevent.

• CVE-2025-71352HIGH 8.1

  picklescan is a security tool designed to inspect Python pickle files for malicious code before they are deserialized. A vulnerability in picklescan before version 0.0.29 allows attackers to bypass its detection mechanisms by using the built-in Python trace.Trace.runctx function within pickle reduce methods. When a user loads a specially crafted malicious pickle file, the code executes without triggering picklescan's safety checks, potentially compromising systems that rely on picklescan for pickle validation.

• CVE-2025-71373HIGH 8.1

  picklescan is a Python security library designed to scan pickle files for malicious code before they are deserialized. A flaw in versions before 0.0.33 allows attackers to craft pickle payloads that bypass picklescan's safety checks by using the operator.methodcaller function—a standard Python function that picklescan fails to recognize as dangerous. When an application loads such a malicious pickle file after passing it through picklescan, arbitrary code execution occurs, defeating the entire purpose of the validation step.

• CVE-2026-11170HIGH 8.1

  Google Chrome on Linux contains a vulnerability in its Chromoting feature (the remote desktop capability) that allows an attacker on the network to gain administrative privileges on your system without needing to interact with you. The vulnerability exists in Chrome versions before 149.0.7827.53. While the Chromium project rates this as medium severity, the actual impact—unauthenticated remote privilege escalation—warrants a CVSS score of 8.1 (HIGH), reflecting the seriousness for Linux desktop environments where Chromoting might be enabled.

• CVE-2026-0097HIGH 8.0

  A logic error in Android's Bluetooth Low Energy (LE) pairing mechanism allows an attacker within wireless range to pair a device without requiring user approval or interaction. An attacker with local access to a Bluetooth-enabled Android device can escalate privileges by circumventing the normal pairing consent flow, potentially gaining full read, write, and execution access on the target device. This is particularly dangerous because it requires no user action to exploit.

• CVE-2026-45588HIGH 7.9

  A weakness in Windows Secure Boot—a fundamental protection mechanism that ensures only authorized code runs during system startup—has been discovered that allows a high-privileged attacker to bypass this security feature on affected systems. The flaw affects multiple versions of Windows 10, Windows 11, and Windows Server across generations. While exploitation requires high-level access locally (such as from an administrator account), a successful bypass could allow an attacker to circumvent boot-time protections, potentially enabling persistence or further system compromise.

• CVE-2026-47656HIGH 7.9

  CVE-2026-47656 is a high-severity vulnerability in the Windows Boot Manager that allows an attacker with administrator privileges to disable or circumvent a security protection mechanism on a local system. While the attacker must already have elevated permissions, once exploited, they can interfere with boot-level security controls—a critical concern because these controls are meant to prevent unauthorized system modifications. This is not currently listed as actively exploited in the wild, but it warrants prompt patching given the scope of affected Windows versions.

• CVE-2026-48568HIGH 7.9

  CVE-2026-48568 is a high-severity vulnerability in Windows Secure Boot that allows a user with administrator or higher privileges to bypass a key security mechanism on their local system. Secure Boot is designed to prevent unauthorized code from running during system startup. This flaw breaks that protection, potentially enabling an attacker with elevated local access to load malicious code earlier in the boot process than normal security controls would permit. The vulnerability affects a broad range of Windows 10 and Windows 11 versions, as well as Windows Server 2012 through 2025.

• CVE-2026-48570HIGH 7.9

  A flaw in Windows Secure Boot allows a high-privileged local attacker to disable or circumvent this critical platform security feature. Secure Boot is designed to prevent unauthorized code from running during system startup; this vulnerability defeats that protection. The attacker needs administrative or system-level access to exploit it, and the impact spans confidentiality and integrity of the entire system. This is not currently known to be exploited in the wild, but it represents a significant risk for any environment where an insider or compromised administrator poses a threat.

• CVE-2026-48575HIGH 7.9

  Windows Secure Boot, a critical firmware security feature designed to prevent unauthorized code execution during system startup, contains a protection mechanism failure that an authorized attacker can exploit locally. An attacker with administrative privileges can bypass Secure Boot controls, potentially allowing them to load unauthorized code at the lowest system level. The vulnerability affects multiple versions of Windows 10 and Windows 11, as well as Windows Server editions spanning 2012 through 2025.

• CVE-2025-48649HIGH 7.8

  CVE-2025-48649 is a local privilege escalation vulnerability affecting Google Android in which an attacker with limited user privileges can reset user-selected permission settings, effectively bypassing the permissions model that Android uses to protect sensitive device capabilities. Because no additional privileges are needed and user interaction is not required, any application with basic local access can trigger this issue to gain unauthorized access to protected device functions—a significant departure from Android's intended permission architecture.

• CVE-2025-48652HIGH 7.8

  A logic flaw in Android's application installation validation code allows a local attacker to bypass Mobile Device Management (MDM) security policies. An attacker with local access to the device can exploit this vulnerability to gain elevated privileges without requiring additional permissions or user interaction. MDM policies are a key security control for organizations managing corporate Android devices, making this bypass a significant concern for enterprise environments.

• CVE-2026-0045HIGH 7.8

  A logic error in Android's Bluetooth RFCOMM connection handling allows a local attacker to bypass the bonding requirement for secure connections. An attacker with local access can escalate privileges without needing special permissions or user interaction, potentially gaining full control over sensitive device functions protected by Bluetooth pairing.

• CVE-2026-0077HIGH 7.8

  CVE-2026-0077 is a privilege escalation vulnerability in Android's ActivityRecord component that allows a local attacker with limited user privileges to launch background applications and gain elevated system access. The flaw stems from a logic error in the resumeConfigurationDispatch function that fails to properly validate or constrain application launch permissions. No special privileges or user interaction are required for exploitation, making this a straightforward attack vector for any app running on an affected device.

• CVE-2026-0087HIGH 7.8

  A logic error in Android's domain verification service allows a local attacker to hijack app links associated with arbitrary applications. By exploiting this flaw, an attacker can redirect app links to malicious apps, potentially intercepting sensitive user actions or data. The vulnerability requires local access but no special permissions or user interaction, making it a meaningful escalation path on compromised or personally-owned devices.

• CVE-2026-0278HIGH 7.8

  A local user on Windows can circumvent data loss prevention (DLP) controls in Palo Alto Networks' Prisma Access Agent by exploiting multiple protection mechanism failures. An attacker with local access can bypass the DLP policies meant to prevent sensitive data from leaving the system. This vulnerability requires local access and does not affect the macOS version of Prisma Access Agent.

• CVE-2026-12214HIGH 7.8

  Qihoo 360 Total Security version 6.0 contains a vulnerability in its Nucleus Engine Monitoring Logic that allows a local attacker with regular user privileges to bypass security protections by manipulating how the system handles network address parameters. This flaw has been publicly disclosed with working exploit code available, creating immediate risk for organizations running this software.

• CVE-2026-45656HIGH 7.8

  A flaw in Windows UEFI firmware allows someone with local access to bypass a built-in security protection. The attacker must already have basic user privileges on the system, but once exploited, they can read sensitive data, alter system files, or disable critical functions. This is a protection mechanism failure—think of it as a lock that should prevent unauthorized actions but doesn't work correctly under certain conditions.

• CVE-2026-61437HIGH 7.8

  PraisonAI, a Python framework for building agent-based workflows, contains a code execution vulnerability in versions before 1.6.78. When processing workflow files that reference Pydantic models as strings, the framework automatically loads a sibling tools.py file from the same directory without proper security checks. An attacker who can place or modify both the workflow configuration and tools.py file can inject malicious Python code that executes with full privileges when the workflow runs. This is a local attack vector requiring file system access, but the impact is severe—complete code execution in the context of the application using PraisonAI.

• CVE-2026-54013HIGH 7.6

  Open WebUI is an offline-capable AI platform that stores user and model metadata, including profile images. A security fix applied to user and webhook profile images in version 0.9.6 was incompletely implemented—model profile images were overlooked. Any authenticated user with default workspace permissions can upload a malicious SVG image disguised as a model's profile picture. When another user views that image, the embedded JavaScript executes in their browser with full account privileges, enabling complete account takeover. The vulnerability requires an authenticated attacker and user interaction, but the permissions involved are enabled by default.

• CVE-2026-14409HIGH 7.5

  A vulnerability in Chrome's V8 JavaScript engine allows attackers to run malicious code within the browser's sandbox if a user performs specific interactions with a crafted webpage. While the vulnerability requires user engagement and is marked as 'Low' severity by Google's internal classification, the CVSS score of 7.5 reflects the potential for complete compromise of user data within the browser process. All Chrome versions before 150.0.7871.46 are affected.

• CVE-2026-55487HIGH 7.5

  pnpm, a widely-used package manager, contains a vulnerability in how it normalizes package source identifiers. Before versions 10.34.2 and 11.5.3, the system would strip parenthesized text from git URLs, tarballs, file paths, and other package sources when checking if they were approved. An attacker could exploit this by crafting a malicious package source that normalizes to match an already-approved source, potentially allowing installation of unauthorized code. The vulnerability requires user interaction (such as running an install command) to trigger, but successful exploitation could grant an attacker significant control over the dependency supply chain.

• CVE-2026-57281HIGH 7.5

  Jenkins Script Security Plugin versions up to 1402.v94c9ce464861 contain a sandbox escape vulnerability. The plugin is designed to safely execute Groovy scripts within a restricted environment, but it fails to properly validate certain Groovy language features called AST transformation annotations. An attacker who can submit sandboxed Groovy scripts to a Jenkins instance can craft malicious code that uses these annotations to break out of the sandbox and execute arbitrary commands with the privileges of the Jenkins process. This requires the attacker to already have legitimate script execution permissions and relies on specific script libraries being available, but the impact—full code execution outside sandbox controls—is severe.

• CVE-2026-48546HIGH 7.3

  KanaDojo, a Node.js application, contains a critical vulnerability in its automated pull request response workflow that allows attackers to escape the intended sandbox isolation and run arbitrary code on GitHub Actions runners. An attacker who can submit a pull request can craft a malicious modification to trigger code execution with the privileges of the automation token, potentially compromising the repository and downstream systems.

• CVE-2026-13601HIGH 7.1

  Yelp, a GNOME help viewer, contains a flaw in how it handles security policies that allow malicious applications running in a Flatpak sandbox to access files they shouldn't. By crafting specially formatted help content containing untrusted stylesheets embedded in SVG documents, an attacker can trick Yelp into reading arbitrary files from the host system and sending that data back over the internet. This bypasses Flatpak's sandbox protections, potentially exposing sensitive user data.

• CVE-2026-14440MEDIUM 6.8

  Cloudflare's Universal SSL feature automatically manages DNS Certificate Authority Authorization (CAA) records to enable TLS certificate issuance on behalf of customers. However, the auto-managed CAA records are intentionally permissive and do not honor stricter security parameters (RFC 8657 accounturi and validationmethods) that customers may set independently. This creates a gap: if a customer configures advanced CAA protections to bind certificate issuance to specific ACME accounts or validation methods, Cloudflare's authoritative DNS overrides those settings with a more permissive record at query time. A sophisticated attacker with access to an ACME account at one of the listed Certificate Authorities could potentially obtain a valid browser-trusted TLS certificate for the affected domain, enabling man-in-the-middle attacks. Exploitation is difficult in practice, requiring the attacker to satisfy domain validation from multiple geographically distributed CA vantage points and work around Cloudflare's global anycast network architecture.

• CVE-2026-10944MEDIUM 6.5

  A flaw in Google Chrome's autofill feature on iOS could allow an attacker to trick a user into visiting a malicious webpage that extracts sensitive information you've saved in your browser—such as payment details, addresses, or credentials—from other websites you use. The vulnerability requires user interaction (visiting the malicious page) but does not require special system permissions or unusual browser configurations to exploit.

• CVE-2026-10950MEDIUM 6.5

  Google Chrome on iOS has a flaw in how it enforces security policies for the autofill feature. An attacker can trick a user into visiting a specially crafted webpage that leaks sensitive data from other websites the user has visited or logged into. The vulnerability requires user interaction (clicking or visiting a malicious link) but doesn't require any special browser configuration or authentication bypass. It affects Chrome versions before 149.0.7827.53 on iOS devices.

• CVE-2026-11206MEDIUM 6.5

  A vulnerability in Google Chrome's Service Worker implementation allows attackers to access sensitive data from websites you visit, even data that should be restricted to specific origins. An attacker can craft a malicious web page that, when visited, exploits insufficient policy checks to leak cross-origin information. This requires user interaction—you must visit the attacker's page—but once there, the browser's protections are bypassed without additional warnings or user awareness.

• CVE-2026-11263MEDIUM 6.5

  A flaw in Google Chrome's WebAuthentication implementation on Android before version 149.0.7827.53 could allow an attacker who has already compromised Chrome's rendering engine to steal sensitive data from websites across different origins. The attacker would need to trick a user into visiting a malicious webpage, but the underlying issue stems from insufficient enforcement of security policies that should prevent cross-origin data leakage. This is a moderate-severity issue because it requires a prior compromise of the renderer process, limiting the attack surface to scenarios where other vulnerabilities or system weaknesses have already been exploited.

• CVE-2026-11288MEDIUM 6.5

  A vulnerability in Google Chrome's CSS handling allows an attacker to leak data from websites you visit to a different origin through a malicious webpage. The flaw stems from insufficient enforcement of browser security policies that normally prevent one website from accessing information from another. An attacker would need to trick you into visiting their crafted page while you're logged into or actively using other sites, but no special browser configuration or advanced user interaction is required beyond a standard click. This is a medium-severity issue affecting multiple operating systems through Chrome.

• CVE-2026-12302MEDIUM 6.5

  A security bypass vulnerability exists in Firefox and Thunderbird's DOM security component that could allow an attacker to circumvent built-in protections. The vulnerability requires no user interaction and can be exploited over the network, though it is limited to information disclosure and integrity impacts without causing system unavailability. Mozilla has addressed this in recent versions of both browser products.

• CVE-2026-13862MEDIUM 6.5

  Google Chrome on iOS has a flaw in how it enforces security policies for Web Authentication features like passkeys and security keys. An attacker positioned on the same network as a victim could craft a malicious webpage that tricks the browser into leaking sensitive data from other websites the user has visited. The issue requires the attacker to be on a privileged network position and requires user interaction to click on a malicious link, but if successful can expose confidential information across website boundaries.

• CVE-2026-13876MEDIUM 6.5

  Google Chrome versions before 150.0.7871.47 contain a flaw in how the browser handles network traffic that allows an attacker positioned to intercept network communications to bypass the browser's Content Security Policy (CSP) protections. An attacker exploiting this would need to be in a position to monitor or modify traffic between a user and the websites they visit, such as on a shared or compromised network. The attacker cannot directly steal data or crash the browser, but can bypass CSP rules that normally prevent malicious scripts from running, potentially enabling further attacks if the attacker can inject their own content.

• CVE-2026-13886MEDIUM 6.5

  Google Chrome versions prior to 150.0.7871.47 contain a flaw in how Isolated Web Apps enforce content security policies. An attacker can craft a malicious HTML page that, when visited by a user, bypasses these protections—potentially allowing unauthorized modifications to web content or application behavior. The vulnerability requires user interaction (clicking a link or visiting a page) but does not require the victim to be logged in or have special privileges.

• CVE-2026-13904MEDIUM 6.5

  Google Chrome on iOS has a flaw in its Safe Browsing feature that allows attackers to bypass navigation restrictions designed to protect users from malicious sites. An attacker could craft a malicious HTML page that, when visited, tricks Chrome's safety mechanisms into allowing navigation to a blocked site. The vulnerability requires user interaction—the user must visit the attacker's page—but does not require special system privileges or browser configuration.

• CVE-2026-13910MEDIUM 6.5

  Google Chrome on Android contains a vulnerability in its WebXR implementation that can allow attackers to steal data from websites you visit. An attacker would need to trick you into visiting a malicious webpage, but once there, they could potentially access sensitive information from other websites you have open in your browser. The vulnerability affects Chrome versions before 150.0.7871.47 on Android devices.

• CVE-2026-14050MEDIUM 6.5

  A weakness in how Google Chrome enforces security policies for the Passwords feature before version 150.0.7871.47 could allow an attacker to trick users into visiting a malicious webpage that leaks sensitive data across different websites. The attacker cannot directly compromise the browser; instead, they must craft a convincing HTML page and convince the user to visit it. Once a user is on that page, the vulnerability permits unauthorized access to information from other origins—effectively bypassing the browser's same-origin policy protections.

• CVE-2026-14059MEDIUM 6.5

  A security weakness in Google Chrome's Related-Website-Sets feature allows attackers to trick users into visiting a malicious webpage that can steal data from other websites the user is logged into. The vulnerability affects Chrome versions before 150.0.7871.47 and requires user interaction (clicking a link or visiting the malicious page). While the Chromium project rated this as low severity, the CVSS score reflects a medium-risk profile because of its potential to expose sensitive cross-origin information without the user's knowledge.

• CVE-2026-59207MEDIUM 6.5

  n8n's AI Agents feature contains a credential access control bypass that allows a member-level user with read-only access to a shared credential to exfiltrate that credential's secret to an attacker-controlled server. The vulnerability exists because the platform fails to enforce domain restrictions on HTTP requests made by MCP (Model Context Protocol) tools used within AI Agents workflows. An attacker with legitimate but limited access to a credential can point an MCP tool at their own server and capture the credential secret during execution. This affects n8n versions prior to 2.27.4 and 2.28.1.

• CVE-2026-10174MEDIUM 6.3

  Aider-AI's Aider version 0.86.3 contains a flaw in how it processes pre-commit hook arguments. An attacker with valid credentials can manipulate the git-commit-verify argument to bypass security protections that normally prevent unauthorized code commits. The vulnerability requires network access and prior authentication, making it a concern primarily for development teams using Aider in shared or untrusted environments. Public exploit code exists, increasing the practical risk.

• CVE-2026-14625MEDIUM 6.3

  NousResearch's hermes-agent, a tool for managing AI agent workflows, contains a vulnerability in how it handles shell execution commands. An authenticated attacker can bypass security controls that normally prevent dangerous operations, potentially gaining the ability to execute arbitrary commands on the affected system. The flaw exists in versions up to 0.15.2 and has already been disclosed publicly with working exploit code available, making it an active risk for organizations using vulnerable deployments.

• CVE-2025-30431MEDIUM 5.5

  CVE-2025-30431 is a medium-severity vulnerability in Apple macOS that allows a malicious application already running on a user's computer to access private information. The flaw stems from inadequate validation checks in the operating system. Because an attacker must first get a malicious app onto the system and have it execute with user-level privileges, the real-world risk depends heavily on how the app gets installed—whether through social engineering, supply-chain compromise, or user mistake. Apple has patched this across three recent macOS versions.

• CVE-2026-45595MEDIUM 5.4

  Windows Mark of the Web (MOTW) is a security feature that warns users when they open files downloaded from the internet. CVE-2026-45595 is a flaw that allows attackers to bypass this warning through network-based attacks. An attacker could craft a malicious file that appears safe to the operating system, potentially tricking users into executing it without the usual security prompts. While the vulnerability requires user interaction to be exploited, it undermines a foundational defense mechanism in Windows.

• CVE-2026-11174MEDIUM 5.3

  CVE-2026-11174 is a Site Isolation bypass vulnerability in Google Chrome that allows an attacker who has already compromised a browser's renderer process to read sensitive data from other websites. The flaw stems from improper implementation of Chrome's Site Isolation feature, a critical security boundary that normally prevents one website from accessing another's data. An attacker would need to first gain control of the renderer process—typically through a separate vulnerability—and then use a specially crafted HTML page to circumvent this protection. While the attack requires a prior compromise, the consequence of success is confidentiality loss across site boundaries.

• CVE-2026-44646MEDIUM 5.3

  LiquidJS, a JavaScript template engine compatible with Shopify and GitHub Pages, has a flaw in how it handles security settings when rendering partial templates using the {% render %} tag. When a developer configures LiquidJS with a permissive default setting but then tries to tighten security for specific untrusted template content, that tightened security is silently ignored for any partials rendered within it. This allows an attacker to access properties from the JavaScript prototype chain—essentially reading unintended internal object data—even when the developer believed they had locked down access.

• CVE-2026-45655MEDIUM 5.3

  Windows BitLocker, Microsoft's full-disk encryption feature, contains a flaw that allows an attacker with physical access to a device to bypass its protection and access encrypted data. The vulnerability affects multiple versions of Windows 10, Windows 11, and Windows Server. While an attacker must have hands-on access to the machine, the risk is significant because BitLocker is often the last defense against data theft when a device is lost, stolen, or accessed by an insider. This is not a remote attack; it requires physical presence.

• CVE-2026-47676MEDIUM 5.3

  Hono, a JavaScript web framework, contains a path handling vulnerability in versions before 4.12.21 that affects how mounted sub-applications receive requests. When URLs contain percent-encoded characters (like %C3%A9 for é), the framework strips the mount prefix incorrectly, causing the sub-application to see a mangled path. This can lead to requests being routed to unintended endpoints or exposing sensitive information through path confusion.

• CVE-2026-53949MEDIUM 5.3

  Ghost, a popular Node.js-based content management system, contains a flaw in how it validates filter parameters on public API endpoints. Attackers can exploit this weakness to bypass security checks and access private fields that should be hidden, potentially including user password hashes. The severity of the exposure depends on the underlying database: SQLite instances leak complete password hashes, while MySQL instances lose case sensitivity information in hashes, which significantly hampers offline brute-force feasibility. The vulnerability affects Ghost versions 5.46.1 through 6.21.2, and the issue is resolved in version 6.21.2.

• CVE-2026-60086MEDIUM 5.3

  PraisonAI versions before 4.6.78 have a weakness in their prompt injection defense system. The defense is designed to block dangerous attacks, but it only stops threats it classifies as CRITICAL—which requires matching three or more threat detection families at once. Attackers can bypass this by crafting simpler prompt injections that are classified as HIGH severity instead, allowing malicious inputs to reach the underlying AI model without being blocked.

• CVE-2026-49859MEDIUM 5.2

  Deno, a modern JavaScript and TypeScript runtime, contains a network access control bypass in versions before 2.8.1. When a script uses fetch() to make network requests, Deno's security model allows administrators to block connections to specific IP addresses or networks using the --deny-net flag. However, the vulnerability exists because Deno only checked the hostname against the deny list but failed to verify the actual IP addresses that hostname resolved to. An attacker could craft a malicious domain name that appears to pass the hostname restrictions yet resolves to a blocked IP address, allowing unauthorized network access. This was patched in version 2.8.1.

• CVE-2026-59854MEDIUM 4.9

  SiYuan is a popular open-source personal knowledge management platform. Versions prior to 3.7.1 contain a file-access vulnerability in the global file copy API endpoint that allows authenticated users with administrator privileges or valid API tokens to extract sensitive credential files from the host system. An attacker can exploit a gap in the file-path validation logic to copy commonly-used credential stores—such as Git credentials, SSH keys, database passwords, Kubernetes configuration, and Docker authentication tokens—from the user's home directory into the SiYuan workspace, then download them via the file API. The vulnerability requires valid authentication, limiting exposure to insider threats or compromised service accounts.

• CVE-2026-49316MEDIUM 4.6

  A vulnerability in the 2025 Indian Motorcycle Scout Bobber + Tech model allows someone with access to the motorcycle's wireless network to disable anti-theft protections and operate the vehicle without proper authorization. An attacker can manipulate error messages on the motorcycle's internal communication system (CAN bus) to silence the Wireless Control Module, which normally enforces shutdown commands tied to the immobilizer. Once this module stops communicating, other systems on the motorcycle treat its silence as normal rather than a security event, leaving the bike vulnerable to theft despite the immobilizer lock never being engaged.

• CVE-2026-49325MEDIUM 4.6

  Indian Motorcycle's 2025 Scout Bobber + Tech model contains a physical security flaw in its anti-theft system. An attacker with access to the motorcycle's Wireless Control Module (WCM) wiring harness can disconnect a specific wire pair to bypass the PIN-protected shutdown mechanism, leaving the bike fully operational and vulnerable to theft. The vulnerability exploits a gap in how the motorcycle's engine control unit (ECU) validates shutdown signals—it cannot tell the difference between a legitimate shutdown command and a severed wire.

• CVE-2025-48571MEDIUM 4.3

  A logic error in Android's Bluetooth management code (btm_sec.cc) creates a window where attackers can intercept incoming SMS messages. An attacker on the network could exploit this to read message content without needing special device permissions, but requires user interaction to succeed—for instance, the user must be engaged in a Bluetooth operation when the interception occurs. The vulnerability carries medium severity; while SMS interception is serious, the requirement for user interaction and network proximity limits the attack surface.

• CVE-2026-11219MEDIUM 4.3

  Google Chrome versions before 149.0.7827.53 contain a flaw in how the browser implements navigation controls. An attacker can craft a malicious HTML page that, when visited, bypasses intended navigation restrictions—essentially allowing the page to navigate the browser or access certain destinations in ways it shouldn't be able to. The attack requires user interaction (clicking or visiting the page), but no special browser privileges. While Chromium rates this as Low severity internally, the CVSS scoring reflects Medium severity due to the potential for integrity compromise through navigation spoofing.

• CVE-2026-11234MEDIUM 4.3

  Google Chrome versions before 149.0.7827.53 contain a vulnerability in the FoldableAPIs feature that allows a remote attacker to bypass site isolation—Chrome's core security boundary that separates web pages from each other—if the attacker has already compromised the renderer process. Site isolation is one of Chrome's strongest defenses against malicious websites stealing data from other tabs or extensions. This vulnerability requires both a compromised renderer and user interaction, limiting the immediate threat but warranting timely patching.

• CVE-2026-11260MEDIUM 4.3

  Google Chrome versions before 149.0.7827.53 contain a flaw in how it handles permissions that allows attackers to bypass the browser's Content Security Policy (CSP) protections via a specially crafted webpage. While the underlying browser vulnerability severity is rated as low, the CVSS assessment elevates this to medium risk because it requires user interaction but could enable an attacker to execute unintended behavior or inject content that CSP should block. The issue affects Chrome on Windows, macOS, and Linux.

• CVE-2026-11264MEDIUM 4.3

  Google Chrome versions before 149.0.7827.53 contain a flaw in how Content Security Policy (CSP) is enforced. An attacker can craft a malicious HTML page that, when visited by a user, bypasses the browser's CSP protections. This allows the attacker to inject or execute content that the website owner intended to block, potentially leading to credential theft, session hijacking, or other attacks that degrade site security. The vulnerability requires user interaction—the victim must visit the malicious page—and does not directly compromise the browser itself or enable data exfiltration.

• CVE-2026-11266MEDIUM 4.3

  Google Chrome versions before 149.0.7827.53 contain a flaw in the Safe Browsing feature that allows a remote attacker to bypass its protections by delivering a specially crafted file. An attacker would need to trick a user into opening or interacting with the malicious file, but if successful, the user's safety checks could be circumvented, potentially allowing access to sites or content that Safe Browsing would normally block.

• CVE-2026-11292MEDIUM 4.3

  Google Chrome versions before 149.0.7827.53 contain a flaw in the Blink rendering engine that allows attackers to bypass Content Security Policy (CSP) protections through a specially crafted webpage. An attacker would need to trick a user into visiting a malicious site, where the weakness could enable injection of unintended content or scripts that CSP was supposed to prevent. While Chromium rates this as low severity, the CVSS score reflects moderate impact potential because CSP bypass can lead to unauthorized modifications of page behavior.

• CVE-2026-11695MEDIUM 4.3

  Google Chrome prior to version 149.0.7827.103 contains a flaw in its password handling logic that could allow an attacker to leak sensitive data across website boundaries. An attacker would need to craft a malicious HTML page and convince a user to visit it, but the vulnerability itself does not require the user to take additional actions beyond normal browsing. The leaked data is restricted to information accessible within the browser context of the affected user.

• CVE-2026-14058MEDIUM 4.3

  Google Chrome versions before 150.0.7871.47 contain a policy enforcement gap in its HTML parser that allows remote attackers to circumvent Content Security Policy (CSP) protections. An attacker can craft a malicious HTML page that, when visited by a user, bypasses CSP restrictions that would normally prevent inline scripts, external resource loading, or other potentially dangerous content execution. This is a client-side vulnerability requiring user interaction—a victim must visit the attacker's page for the bypass to occur.

• CVE-2026-14076MEDIUM 4.3

  Google Chrome versions before 150.0.7871.47 contain a weakness in how the browser enforces Content Security Policy (CSP), a security feature that restricts which resources a webpage can load. An attacker could craft a malicious HTML page that tricks the browser into loading content that should have been blocked by CSP rules, potentially allowing injection of unwanted scripts or other resources. The attack requires user interaction—the victim must visit the malicious page—but succeeds against unpatched Chrome installations on Windows, macOS, and Linux.

• CVE-2026-14092MEDIUM 4.3

  Google Chrome versions before 150.0.7871.47 contain a privacy flaw that allows attackers positioned on a network path between a user and servers to intercept and expose data that should remain isolated between different websites. An attacker must trick or socially engineer the user into visiting a malicious page, but once that happens, the browser's normal cross-origin protections can be bypassed through crafted network traffic. The issue affects Chrome on Windows, macOS, and Linux.

• CVE-2026-53845MEDIUM 4.3

  OpenClaw versions before 2026.5.6 contain a vulnerability that allows authenticated users to bypass security hooks designed to audit and enforce policies on skill command execution. When skill commands are routed through a specific dispatch path, they skip the before-tool-call hooks that normally intercept and validate these operations. An attacker with valid credentials could exploit this to execute commands without triggering audit logs or policy checks.

• CVE-2026-59223MEDIUM 4.3

  Open WebUI versions before 0.10.0 contain a flaw in how they validate domain blocklists used to control which external URLs can be fetched by the platform. An authenticated attacker can bypass these restrictions by manipulating the URL path or exploiting sibling-domain matching logic, potentially gaining access to internally restricted resources. The fix is available in version 0.10.0 and later.

• CVE-2026-12457MEDIUM 4.2

  Google Chrome versions prior to 149.0.7827.155 contain a flaw in how extensions are implemented that allows an attacker who has already compromised Chrome's renderer process to escape the site isolation sandbox and access content from different websites. Site isolation is Chrome's core defense that prevents malicious code running on one site from stealing data from another. This vulnerability requires the attacker to have already gained code execution in the renderer—a significant prerequisite—but if achieved, it undermines that critical isolation boundary.

• CVE-2025-52609LOW 3.7

  HCL iControl is missing HTTP security headers that would instruct modern web browsers to block cross-site scripting (XSS) attacks. Without these headers—such as Content-Security-Policy or X-XSS-Protection—the application relies on older browser XSS filters that are inconsistently implemented and increasingly deprecated. An attacker could craft malicious input that, when processed by iControl, gets reflected in responses without proper sanitization, potentially allowing script execution in users' browsers.

• CVE-2026-3472LOW 3.5

  Mattermost chat server has a vulnerability where AI bot result posts bypass markdown image rendering restrictions. An authenticated user can craft specially formatted content that, when viewed by a victim in their Mattermost client, triggers image requests to an attacker's server—potentially exfiltrating data about what the victim is viewing. The vulnerability requires a valid Mattermost account and user interaction (clicking or viewing the malicious post), limiting but not eliminating risk in collaborative environments.

• CVE-2026-45459LOW 3.3

  Microsoft Excel has a flaw that allows someone with local access to bypass a built-in security protection mechanism. An attacker would need to trick a user into opening a specially crafted Excel file on their machine. The vulnerability exposes some information (such as file contents or formulas) but cannot be used to modify data or crash the application. This is a low-risk issue with limited real-world impact.

• CVE-2026-58052LOW 3.3

  7-Zip for Windows has a flaw in how it handles specially crafted RAR5 archive files. When extracting such an archive, 7-Zip fails to properly maintain the 'Mark-of-the-Web' (MotW)—a security marker that Windows uses to warn users about potentially unsafe files downloaded from the internet. An attacker can exploit this by embedding malicious stream records in a RAR5 file that bypass 7-Zip's protection and allow the tool to extract content that appears safe to Windows, defeating SmartScreen warnings and potentially tricking users into running malware.

• CVE-2026-11247LOW 3.1

  A flaw in Google Chrome's CustomTabs feature on Android allows an attacker to leak data across website boundaries through a specially crafted webpage. The vulnerability requires user interaction and is difficult to exploit, affecting Android devices running Chrome versions before 149.0.7827.53. While the risk is low, it represents a potential privacy leak in a widely used mobile browser component.

• CVE-2026-11684LOW 3.1

  A policy enforcement gap in Google Chrome's Network component allowed attackers who had already compromised Chrome's utility process to steal cross-origin data by serving a specially crafted HTML page. This is a post-compromise attack where the attacker has already gained some level of access to the browser process itself, then exploits this weakness to read data that should be isolated between different websites.