Security news

Microsoft has released two open-source tools, Rampart and Clarity, to help developers secure AI agents. The tools are designed for safety verification during early development, addressing risks as AI agents gain more operational authority. This is part of Microsoft's push for continuous AI safety engineering.

Standard cloud cost-saving practices, like downsizing underused GPUs, don't apply to secure AI training. The usual utilization metrics can be misleading for these specialized workloads, creating a blind spot for FinOps teams and leading to incorrect infrastructure decisions.

A new analysis highlights how advanced AI models can now autonomously discover software vulnerabilities at an unprecedented scale. This fundamentally changes the balance between offense and defense in cybersecurity, making traditional disclosure timelines potentially obsolete and requiring an urgent rethinking of security strategies.

Apple released its annual fraud prevention report, stating it stopped over $2.2 billion in potentially fraudulent transactions last year. The company uses this data to highlight the value of its curated App Store, arguing its security measures and review process protect both users and developers.

In a recent discussion, experts from Dataiku and 1Password explored the next frontier of AI challenges. They covered the essentials of data governance, managing complex data supply chains, and the critical need for robust security frameworks to protect increasingly autonomous and interconnected AI agent swarms.

Snowflake's platform on Google Cloud in Melbourne has successfully completed an IRAP assessment, meeting the Australian Government's 'Protected' security standard. This certification allows public sector agencies and regulated industries in Australia to use the data platform for handling sensitive information and critical workloads on GCP.

HashiCorp Vault has introduced support for SCIM, a standard protocol for automating user identity management. This update enables enterprises to automatically provision and manage users and groups in Vault, ensuring consistency with their primary identity providers. It simplifies access control, enhances security, and improves operational efficiency.

Elastic has released version 9.4.2 of the Elastic Stack. This is a security-focused update that addresses potential vulnerabilities found in previous versions. All users are strongly encouraged to upgrade their deployments to this latest version to ensure their systems remain secure and protected.

Microsoft has updated Azure Logic Apps with sandboxed code interpreters. This allows AI agents within workflows to safely generate and execute Python, JavaScript, C#, and PowerShell code, positioning Logic Apps as a platform for building AI-powered integrations.

A new AI coding agent named Claw-Coder runs entirely on a local machine, addressing privacy and security concerns associated with cloud-based models. It uses Retrieval-Augmented Generation (RAG) and knowledge graphs to enhance the performance of smaller, local language models, offering a private alternative to tools like Codex.

Microsoft is highlighting the security challenges organizations face when adopting AI. As companies like St. Luke’s and ManpowerGroup move to AI-powered models, establishing a strong security foundation for cloud, data, and identity becomes essential for enabling responsible innovation and managing new risks.

Snyk has launched Evo Continuous Offensive Security, a new AI-powered product for penetration testing. It aims to continuously find and fix vulnerabilities in applications, specifically addressing the security challenges posed by the rapid development and deployment of AI-generated code.

The Kubernetes project is correcting historical CVE records for some older, unfixed vulnerabilities. This update will cause vulnerability scanners to flag new issues that were previously miscategorized. Teams should prepare to reassess their security posture based on the updated data before 2026.

GitLab has released version 19.0, a major update focused on strengthening its DevSecOps platform. The release aims to provide a more coordinated and comprehensive toolset for software development, security, and operations, with key enhancements to secrets management.

Researchers have developed a new attack called CrossMPI that targets multimodal AI models. It uses nearly invisible changes in images to manipulate how the AI interprets both visual and text inputs. This technique bypasses safety measures without altering the original text prompt, creating new security risks.

Employees are increasingly using unapproved AI tools, creating security risks. A new guide outlines a five-step approach for companies to manage this "shadow AI." The goal is to establish practical governance and security policies without hindering employee productivity or innovation, balancing control with enablement for teams.

Cryptologist Bruce Schneier noted that artist Laurie Anderson has quoted him in her new album and in interviews. The quote, "If you think technology will solve your problems, you don't understand technology and you don't understand your problems," highlights a critical perspective on technology's role.

Cloudflare tested Anthropic's security-focused AI model, Mythos Preview, on its own infrastructure. As part of Project Glasswing, they used the model to scan over fifty internal code repositories to identify potential vulnerabilities and understand how attackers might leverage similar advanced AI tools in the future.

A practical guide explains the networking concept of Virtual Local Area Networks (VLANs). It details how VLANs work by dividing a single physical network into multiple, isolated virtual segments. This is crucial for improving security, managing traffic, and organizing devices without needing separate physical hardware.

A popular GitHub Actions workflow, `actions-cool/issues-helper`, has been compromised in a supply chain attack. Attackers altered repository tags to point to malicious code designed to steal sensitive credentials from CI/CD environments and send them to an external server.

Ubuntu has released a security update for the Linux kernel used in NVIDIA Tegra systems. The update addresses several vulnerabilities in network drivers, NVME drivers, and Netfilter that could allow an attacker to compromise an affected system. Applying the patch is recommended for all users.

The US Cybersecurity and Infrastructure Agency (CISA) accidentally exposed sensitive credentials, including plaintext passwords and SSH keys, in a public GitHub repository. The repository, discovered by security firm GitGuardian, was accessible for an extended period before being taken offline after a security researcher reported the issue.

Ubuntu has patched multiple vulnerabilities in Unbound, a popular DNS resolver. The flaws could allow remote attackers to cause a denial of service or, in some cases, execute arbitrary code. One critical issue affects recent Ubuntu LTS versions, making immediate updates essential for affected systems.

A nine-year-old vulnerability (CVE-2026-46333) has been found in the Linux kernel. The flaw, caused by improper privilege management, allows a local user without special permissions to access sensitive files and execute commands with root privileges on default installations of several major Linux distributions.

A researcher released a proof-of-concept exploit for a new Windows zero-day vulnerability named 'MiniPlasma'. The exploit allows an attacker with local access to gain the highest system privileges on fully patched Windows 10 and 11 machines. Microsoft has not yet released a patch.

A compromised version of the popular Nx Console extension (v18.95.0) was published on the VS Code Marketplace. The malicious version, installed by developers, contained a credential stealer. The extension, with over 2.2 million installations, poses a significant risk to affected users and their organizations.

Multiple vulnerabilities have been found in the GnuTLS library, a core component for secure communications. The flaws could allow attackers to cause a denial of service, access sensitive information, or potentially execute arbitrary code by sending specially crafted data during a DTLS handshake.

Proof-of-concept exploit code, named DirtyDecrypt, has been publicly released for a recently patched Linux kernel vulnerability. The flaw allows for local privilege escalation, enabling an attacker with local access to gain higher-level system permissions. The vulnerability was discovered and reported by security researchers in early May.

The npm registry has experienced another malware attack, this time affecting the AntV data visualization tool. The incident occurred after an attacker compromised the credentials of a maintainer for the popular `timeago.js` library, highlighting ongoing risks in the open-source software supply chain.

A new vulnerability called YellowKey allows attackers with access to a Windows device to bypass Bitlocker encryption and access files. Microsoft is working on a permanent patch for the flaw (CVE-2026-45585) and has released temporary mitigation steps for companies to implement immediately.

A security researcher has released a proof-of-concept exploit for a new Windows zero-day vulnerability called MiniPlasma. The exploit can grant full SYSTEM privileges on fully patched Windows 11 systems. The flaw is reportedly an unfixed version of a bug Microsoft claimed to have patched in 2020.

GitHub is investigating a claim by a threat actor group called TeamPCP. The group alleges it accessed GitHub's internal repositories and is attempting to sell the platform's source code on a cybercrime forum. GitHub has found no evidence that customer data has been impacted so far.

Ubuntu has patched several vulnerabilities in the Linux kernel. One major flaw, known as 'Copy Fail,' could allow a local attacker to escalate privileges or escape a container. Other issues were found in subsystems like networking, cryptography, and I/O, which could also lead to system compromise.

Grafana Labs has disclosed a security incident where attackers used a stolen GitHub access token to access its environment. The breach resulted in the unauthorized download of some of its source code. Grafana is investigating but states no customer data was compromised.

Microsoft has released a mitigation for a BitLocker security bypass vulnerability known as "YellowKey." The zero-day flaw, tracked as CVE-2026-45585, was publicly disclosed last week and carries a CVSS score of 6.8, affecting the Windows disk encryption feature.

Apple announced it has prevented over $11 billion in fraudulent transactions on its App Store since 2018. The company's fraud prevention systems, which combine machine learning with human review, stopped more than $2.2 billion in a single year alone, safeguarding developers and users.

The creator of NanoClaw, a secure, containerized platform for running AI agents, has turned down a $20 million buyout offer. Instead, the company has secured $12 million in a seed funding round to continue developing its sandboxed platform for AI automation and marketing.

Microsoft has uncovered a supply chain attack targeting the @antv npm ecosystem. Attackers compromised a maintainer's account to publish malicious versions of data-visualization packages. The code aims to steal credentials from CI/CD pipelines and affects widely used libraries like echarts-for-react.

A proof-of-concept exploit is now public for a high-severity Linux kernel vulnerability dubbed 'DirtyDecrypt.' The flaw allows a local user to gain root privileges on systems with the `rxgk` module enabled. Major distributions are affected, and patches are available. Teams should update their systems immediately.

Grafana Labs confirmed a security breach limited to its GitHub environment, exposing public and private source code. The company stated that its investigation found no evidence of customer production systems being compromised. The incident was linked to a supply chain attack involving a TanStack npm package.

A contractor's public GitHub repository accidentally exposed sensitive credentials. The leak included access keys for US government AWS accounts and internal systems for the Cybersecurity and Infrastructure Security Agency (CISA). A researcher from GitGuardian discovered the exposure, which was then reported by security journalist Brian Krebs.

GitHub has disclosed a security breach where an attacker gained unauthorized access to its internal repositories. The compromise originated from a malicious third-party VS Code extension on an employee's device. While thousands of internal repos were exfiltrated, GitHub reports no evidence of impact on customer data.

A government contractor's public GitHub repository accidentally exposed credentials for US government AWS accounts and internal CISA systems. The leak, discovered by a security researcher, included sensitive access keys found within the repository's commit history and developer notes, highlighting significant security risks.

Ubuntu has released a security update for the Linux kernel used in Intel IoTG real-time systems. The patch addresses several vulnerabilities in the SMB network file system, Netfilter, and io_uring subsystems. If left unpatched, an attacker could potentially use these flaws to compromise an affected system.

Ubuntu has issued a security notice for multiple Linux kernel vulnerabilities. The update patches several flaws, including a critical issue in a cryptographic module dubbed "Copy Fail." This specific vulnerability could allow a local attacker to gain elevated privileges or potentially escape from a container environment.

A new report highlights that traditional security benchmarks are ineffective for evaluating AI systems. Unlike standard software, AI security is an emergent property that cannot be measured by simple tests, challenging teams to rethink how they approach securing their AI models and applications.

GitHub is updating its bug bounty program to handle a massive increase in low-quality submissions, many generated by AI tools. The company is tightening its standards and will now reward certain types of lower-impact vulnerability reports with merchandise instead of cash payments to discourage spam.

Microsoft has disrupted a "malware-signing-as-a-service" operation run by a group called Fox Tempest. The service abused Microsoft's own code signing system to make malicious software, including ransomware, appear legitimate, compromising thousands of machines and networks globally.

GitHub has confirmed that a recent breach of 3,800 internal repositories was caused by a malicious VS Code extension. The extension was compromised in a wider supply-chain attack targeting the popular TanStack npm packages, highlighting the growing risks of software dependencies.

Microsoft has released two open-source tools, RAMPART and Clarity, to improve the safety of AI agents. As AI systems increasingly perform actions on behalf of users, these tools help developers test for security risks and validate assumptions throughout the development workflow, making agentic AI safer.

Microsoft has released two new open-source tools, RAMPART and Clarity, to help developers improve the security of AI agents. RAMPART is a testing framework designed to identify and measure security risks in AI systems during the development phase, making it easier to build safer applications.

A new report from Orchid Security reveals that 57% of enterprise identities are “identity dark matter”—unseen and unmanaged. This growth in unmanaged access points creates significant security vulnerabilities, especially as companies rapidly adopt Agent AI, which can exploit these gaps.

A new variant of the 'SHub' macOS infostealer is targeting users with fake Apple security update prompts. The malware uses AppleScript to display a convincing dialog box, tricking users into entering their password. This installs a backdoor, giving attackers access to steal browser data and cryptocurrency wallets.

A new exploit called YellowKey can bypass Windows 11's BitLocker full-disk encryption. Published by a security researcher, the attack works on default deployments but requires direct physical access to the target computer. The vulnerability affects systems that use a TPM to store decryption keys.

The SGLang AI framework has three critical vulnerabilities, including two for remote code execution. An attacker with network access can exploit them if the multimodal mode is enabled. The project maintainers have not responded, and no patch is currently available for these significant security flaws.

A new zero-day vulnerability in Microsoft Exchange, identified as CVE-2026-42897, is being actively exploited. The flaw is a cross-site scripting (XSS) issue that allows attackers to compromise Outlook Web Access (OWA) mailboxes. Microsoft has not yet released a patch to fix the vulnerability.

Docker is highlighting critical security failures in the AI coding agent ecosystem. Citing a report that developers use AI in 60% of their work, the company warns that the shift to coordinated agent teams is creating new vulnerabilities for developer infrastructure.

A recently leaked malware kit named "Shai-Hulud" is now being used in a new campaign targeting the npm registry. Attackers are publishing infected packages to steal sensitive information from developers' systems, including credentials and cryptocurrency wallets. This highlights the ongoing risk of software supply chain attacks.

OpenClaw has published its security roadmap, detailing future plans to enhance its open-source AI platform. Key focus areas include securing the software supply chain with SLSA, implementing runtime sandboxing for models, improving data encryption and privacy, and pursuing SOC 2 and ISO 27001 compliance.

Grafana has disclosed a security incident where an unauthorized party gained access to its GitHub environment using a stolen token. The attacker was able to download the company's codebase. Grafana's investigation found no evidence that customer data or systems were affected by the breach.