Since September 15, 2026, the Gheist campaign has been abusing stolen GitHub tokens to commit a malicious workflow named sync-check to accessible repositories. This workflow attempts to print repository secrets into logs, which are then exfiltrated via GitHub's API before the logs are deleted. The campaign has affected over 850 compromised accounts and more than 13,500 planted branches, impacting both personal and enterprise private repositories. GitHub's approval controls have prevented exploitation in public repositories, but private and enterprise repositories remain vulnerable. The campaign operates without a command-and-control server, uses the user agent 'gheist', and has multiple evolving variants. Approximately two-thirds of affected accounts have been removed by GitHub, likely as a response measure.
CVE-2026-85787 is a vulnerability in the SQL validation component of Amazon awslabs postgres-mcp-server versions prior to 1.1.7. It allows an unauthenticated actor to modify data beyond the intended read-only scope by injecting crafted SQL through content submitted by an authenticated user. The issue is fixed in version 1.1.7. Mitigation includes upgrading to version 1.1.7 or later and running the MCP server with a minimal-privilege Postgres role to enforce database-level access controls.
A cross-site scripting (XSS) vulnerability exists in the yahoo serialize-javascript library prior to version 7.1.3. The flaw arises because the function-source escaping logic fails to neutralize a bare closing-script prefix when attacker-controlled function source text ends exactly at that prefix. This can allow an attacker who controls a serialized function's source text to inject malicious markup that becomes live in the page. The issue is fixed in version 7.1.3.
The Arrakis 2026 Autonomous Agents Threat Report reveals that a significant majority of enterprise autonomous agents remain dormant yet retain active credentials, posing security risks. Notably, 76% of agents have not run in 60 days but still have valid credentials and active triggers. Additionally, 24% of dormant agents can be activated externally via unsigned webhooks, and 20% can delegate tasks to other agents with higher permissions, potentially bypassing intended access controls. The report highlights a common operational gap where agent credentials and approvals are not revoked or expired after inactivity, unlike employee access which is promptly removed upon departure.
AI agents with autonomous capabilities can be manipulated through malicious inputs such as emails, websites, and plugins to perform unauthorized actions including executing commands, stealing data, or deleting files using their legitimate permissions. Gen Digital's research demonstrated these risks without exploiting software vulnerabilities, highlighting the need for trust and control mechanisms around AI agents. To address these risks, Gen Digital developed the Agent Trust Hub, which includes skill verification, secure connectivity via an AI-native VPN, runtime safety controls that intercept dangerous actions, and tools to improve data transparency. The platform also promotes open standards for runtime safety and independent testing guidelines to build trust in AI agent security.
CVE-2026-95711 is an authentication bypass vulnerability in the web-based management interface of ArubaOS (AOS) version 10.4.0.0 through 10.8.1.1. This flaw allows an unauthenticated adjacent attacker to bypass authentication controls and execute arbitrary commands on the underlying operating system if certain external conditions are met. The vulnerability has a high severity rating with a CVSS score of 7.1.
CVE-2026-79884 is a high-severity vulnerability in the web-based management interface of ArubaOS (AOS) version 8. It allows an unauthenticated adjacent attacker to bypass authentication controls and access sensitive information in cleartext. This exposure could lead to further compromise of network infrastructure.
CVE-2026-79881 is an authentication bypass vulnerability in the underlying operating system of ArubaOS (AOS) version 8.10.0.0 through 8.10.0.23 and 8.13.0.0 through 8.13.3.0. This flaw allows an unauthenticated remote attacker to bypass authentication controls and execute arbitrary code with privileged access, potentially leading to full system compromise.
CVE-2026-79869 is an authentication bypass vulnerability in ArubaOS (AOS) versions 10.4.0.0 through 10.4.1.13, 10.7.0.0 through 10.7.2.6, and 10.8.0.0 through 10.8.1.1. It allows an unauthenticated adjacent attacker to bypass authentication controls and execute arbitrary commands with privileged access on the underlying operating system. The vulnerability has a critical severity with a CVSS score of 9.6.
CVE-2026-79868 is an authentication bypass vulnerability in ArubaOS (AOS) version 8 that allows an unauthenticated adjacent attacker to bypass authentication controls. Exploiting this flaw could grant administrative access, enabling modification of system configurations and access to sensitive data, resulting in complete system compromise.
CVE-2026-79867 is a critical authentication bypass vulnerability in ArubaOS (AOS) version 8 that allows an unauthenticated adjacent attacker to bypass authentication controls and execute arbitrary commands on the underlying operating system. This leads to complete system compromise. The vulnerability affects specific versions of AOS-8 and has a high CVSS score of 9.6, indicating severe impact on confidentiality, integrity, and availability.
CVE-2026-79855 is a critical authentication bypass vulnerability in the SSH daemon of ArubaOS (AOS) version 8.10.0.0 through 8.10.0.23 and 8.13.0.0 through 8.13.3.0. This flaw allows unauthenticated remote attackers to bypass authentication controls and gain administrative privileges, resulting in complete system compromise of the affected AOS host.
CVE-2026-79850 is a critical vulnerability in ArubaOS (AOS) version 8 that allows unauthenticated remote attackers to bypass authentication controls and execute arbitrary commands as root on the underlying operating system. This leads to complete system compromise. The vulnerability affects specific versions of AOS-8 and has a CVSS score of 10.0, indicating maximum severity.
CVE-2026-75508 is a high-severity vulnerability in graphql-core, a Python port of GraphQL.js. The flaw exists in versions from 3.2.3 up to but not including 3.2.12 and from 3.3.0a3 up to but not including 3.3.0rc1. It involves uncontrolled resource consumption due to the parser's max_tokens limit counting only non-comment tokens, while comment tokens are still retained in memory. This allows an unauthenticated remote attacker to exhaust worker memory by sending a comment-padded GraphQL document before depth or complexity controls are applied. The issue is fixed in versions 3.2.12 and 3.3.0rc1.
A path traversal vulnerability exists in the Kubernetes kubectl client's 'kubectl cp' command on Windows. This flaw allows an attacker who controls container contents to execute arbitrary code and write files to arbitrary paths on the local machine when 'kubectl cp' is used, limited by the local user's system permissions. The issue arises because kubectl runs tar inside the container to create an archive, which is then unpacked locally. This vulnerability affects kubectl clients running on Windows only and has a medium severity rating.
A New Mexico court ordered Meta to pay $25 billion for intentionally deceiving Facebook users about privacy controls, marking the largest fine against the company. The ruling found Meta liable for over 43 million violations of New Mexico consumer protection laws related to misleading statements about user data handling. This case stems from the Cambridge Analytica scandal where Facebook data was harvested for political advertising. Meta disputes the ruling and plans to appeal. The fine significantly exceeds previous penalties, including a $5 billion FTC fine in 2019. The judgment aims to hold tech companies accountable for privacy misrepresentations.
A security vulnerability (CVE-2026-96512) in the sudo utility allows bypass of time-based authorization controls via the TZ environment variable. This affects Red Hat Enterprise Linux 10.0 Extended Update Support versions across multiple architectures. The vulnerability permits users to circumvent NOTBEFORE/NOTAFTER restrictions, potentially enabling unauthorized privileged command execution. Red Hat has issued a security update to address this issue.
CVE-2026-76487 is a vulnerability in Cisco Unified Communications Manager and its Session Management Edition that allows unauthenticated remote attackers to bypass access controls by sending specially crafted HTTP requests. This is caused by improper input validation of certain HTTP requests, potentially enabling attackers to access network areas beyond their authorized scope. The vulnerability affects multiple specific versions of the product. The CVSS score is 5.3, indicating a medium severity level.