This content discusses the competitive disadvantage faced by defenders in cybersecurity due to adversaries using uncensored, abliterated AI language models that automate reconnaissance, phishing, and exploit development. The post highlights the availability of such uncensored models through Silk Compute, which removes refusal behaviors and content filters, enabling more aggressive and unrestricted AI usage. No specific vulnerability or exploit is described.
Vite versions from 6.4.4 up to but not including 6.4.5 have a vulnerability where the development server improperly orders validation and canonicalization of file system access rules. This allows an unauthenticated attacker to bypass server.fs.deny restrictions by appending certain loader query parameters, potentially accessing sensitive files that should be denied. The issue is fixed in version 6.4.5.
Infisical, a secrets management company, has raised $40 million in Series B funding, valuing it at $600 million. The company provides tools for managing secrets, encryption keys, digital certificates, and privileged access across cloud, on-premises, and AI infrastructures. Its Agent Vault technology enables AI agents to access sensitive systems without exposing underlying credentials by injecting secrets at a proxy and logging all requests. Over 1,300 organizations use Infisical's products. The funding will be used to enhance product development and AI agent security capabilities.
CVE-2026-115522 is a vulnerability in Google Agent Development Kit (ADK) for Python that allows an application to access or modify artifacts belonging to another application when both share a FileArtifactService root directory and the same user_id. This occurs because the service omitted the application name when constructing on-disk paths, leading to improper resource isolation. The issue affects versions 1.19.0 through before 1.39.1 and 2.0.0a1 through before 2.7.0. The vulnerability was fixed by enforcing full per-application path scoping in versions 1.39.1 and 2.7.0.
Anthropic, an AI company, disclosed in its IPO filing with the SEC that its AI technologies pose significant risks including potential misuse by bad actors and the possibility of AI systems acting autonomously with harmful consequences. The filing highlights concerns about rapid AI development outpacing safety measures, risks of AI being weaponized or used for harmful surveillance, and operational challenges such as securing compute resources. Anthropic positions its IPO as a step toward mitigating these risks by emphasizing safety and responsible AI development.
Anthropic, an AI company, has filed for a potentially record-breaking IPO while its CEO warns of existential risks posed by advanced AI. The company highlights concerns about rogue AI agents conducting harmful activities and the broader societal risks of AI technology slipping out of human control. The IPO filing includes extensive risk disclosures, including the possibility of AI models being misused for malicious purposes or going rogue. Anthropic advocates for slowing AI development to allow safety measures to catch up and calls for industry-wide and governmental coordination. The company is also involved in a public dispute with the U.S. government over national security concerns related to its technology.
CVE-2026-57187 is a heap-use-after-free vulnerability in Asterisk's PJSIP TCP/SDP handling. It affects multiple versions prior to the fixed releases and requires an authenticated attacker to send a SIP INVITE over a connection-oriented transport and disconnect before a 200 OK response. The issue has been reproduced only with Address Sanitizer, a development tool not used in production. Fixed versions are available.
SQL injection in the database-query custom resource handlers in the Amazon Web Services (AWS) Cloud Development Kit (CDK) @aws-cdk/aws-redshift-alpha module before 2.274.0-alpha.0 could allow a third party who can influence construct property values to run SQL statements with Amazon Redshift superuser privileges, including granting access to all cluster users. The statements are reachable through crafted user names, comments, or passwords.
To remediate this issue, users should upgrade to version 2.274.0-alpha.0 or later and review existing table grants to PUBLIC.
A vulnerability was found in Travisun Opptrix 0.4.0. Affected is the function Buffer.concat of the file /api/channels/webhook/ of the component preParsing handler. Performing a manipulation results in denial of service. The attack can be initiated remotely. The exploit has been made public and could be used. The project has been completely taken offline and deprecated. The project maintainer explains: "The mentioned security risk was a design flaw that manifested primarily in local environments during earlier stages of version development. We previously addressed and resolved this specific risk; however, because the project was restricted to a small learning scope and the main versions were intended to run exclusively on localhost, the security fix was only pushed directly to our GitHub repository before it went private." This vulnerability only affects products that are no longer supported by the maintainer.
A vulnerability has been found in Travisun Opptrix 0.4.0. This impacts an unknown function of the file /api/system-update/import of the component Import Endpoint. Such manipulation leads to code injection. It is possible to launch the attack remotely. The exploit has been disclosed to the public and may be used. The project has been completely taken offline and deprecated. The project maintainer explains: "The mentioned security risk was a design flaw that manifested primarily in local environments during earlier stages of version development. We previously addressed and resolved this specific risk; however, because the project was restricted to a small learning scope and the main versions were intended to run exclusively on localhost, the security fix was only pushed directly to our GitHub repository before it went private." This vulnerability only affects products that are no longer supported by the maintainer.
CVE-2026-115969 is a command injection vulnerability in Travisun Opptrix version 0.4.0 affecting the /api/auth/setup endpoint of the Schedule API component. The flaw allows remote attackers to execute arbitrary commands. The project has been deprecated and taken offline, with no official patches released. The maintainer states the issue was a design flaw mainly in early local development versions and that a fix was pushed privately before the project went private. This vulnerability affects unsupported software and has a medium severity rating with a CVSS 4.0 score of 6.9.
baserCMS is a website development framework. In versions 5.0.0 before 5.3.2 and 5.4.0 before 5.4.2, an authenticated user with only theme-file view permission can traverse outside the theme directory through the optional BcThemeFile plugin's Admin web screen. ThemeFilesController::parseArgs() constructs a path containing literal ../ segments, and the view and edit actions pass it to ThemeFilesService::get() without containment validation. Authorization checks the unresolved request path, after which ThemeFile::_getContents() reads the resolved file and exposes its contents in the form textarea. A client must preserve the dot segments rather than normalize them away. Readable files are limited to .php, .css, .js, and .ctp; this can disclose application source and config/install.php database credentials in MySQL or PostgreSQL deployments, while SQLite configurations have an empty database password. Reading config/app_local.php may expose the Security.salt fallback, which is effective only when not overridden by environment configuration. The flaw does not disclose .env, .pem, or SQLite .db files, does not itself permit file writes or code execution, and does not affect the separate API getFullpath() route. The plugin is disabled by default and unauthenticated requests are not sufficient. This issue has been patched in versions 5.3.2 and 5.4.2.
baserCMS is a website development framework. Starting in version 5.0.0-beta1 and prior to versions 5.3.2 and 5.4.2, `PermissionGroupsService::build()` reads a plugin's `config/permission.php` to decide what each user group may reach. When that file is absent it writes a wildcard allow rule instead: `url` `/baser/<prefix>/<plugin>/*`, `method` `*`, `auth` true, for every non-admin group. BcCustomContent has no such file in any release. After an administrator installs it, the default `operators` group can store and execute arbitrary PHP under the document root. Versions 5.3.2 and 5.4.2 patch the issue.
CVE-2026-107858 is a high-severity vulnerability in Google Agent Development Kit (ADK) for Java versions before 1.13.0. It involves missing origin validation in WebSockets and permissive cross-domain policies, allowing remote attackers to control a locally running agent via malicious web pages. The vulnerability affects the /run_live WebSocket endpoint and REST API, which accept connections and requests from any origin by default. This enables cross-origin attacks that can trigger tools, stream responses, and manipulate sessions and artifacts. The issue is fixed in version 1.13.0 by enforcing same-origin policies and rejecting DNS-rebinding requests over loopback.