Microsoft is updating its identity security framework to address the unique risks posed by autonomous AI agents. Traditional multi-factor authentication is insufficient for these entities because they operate at speeds and scales that bypass human-centric security measures. Consequently, the identity playbook is shifting toward a unified risk model that integrates signals from identity, endpoints, and data layers.

Source

A Ukrainian defense industry official revealed that fully autonomous drones killed Russian soldiers during a specialized battlefield test. The engagement involved ten quadcopters programmed to fly to the front lines before activating a "Terminator mode" to seek and destroy targets independently. This event marks a significant shift in warfare, as the drones operated without any human oversight or active data link during the attack phase.

Source

The ASCILINE Engine is a new open-source tool designed to render high-performance, real-time video streams using pure ASCII characters. By mapping pixels to text-based representations, the software can output 360p video at 30 frames per second within a web browser. The developer claims this method creates a video stream that is difficult to block because it bypasses traditional media filters.

Source

A threat group known as Velvet Ant successfully maintained a presence within a large organization's isolated network for ten years by hijacking the authentication stack. The intrusion began in 2016 when the actors compromised internet-facing systems before pivoting into an air-gapped environment with no direct external connection. By establishing a remote execution path through chained Nginx and FastCGI modifications, the attackers bypassed traditional network segregation without requiring direct internet access.

Source

A coalition of U.S. state attorneys general has launched a broad investigation into OpenAI regarding its business practices and data handling. The ChatGPT creator received a subpoena on Friday requesting documents related to user engagement, advertising, and the management of sensitive consumer and health data. This legal inquiry also examines the company's impact on vulnerable populations, including minors and senior citizens, as well as its internal safety policies.

Source

Organization administrators can now manage GitHub Copilot code review settings at scale through new runner type configurations. These controls allow for the enforcement of specific self-hosted or large runners across all repositories within an organization. By locking these settings, administrators ensure that organizational defaults override individual repository configurations for consistent performance and security.

Source

The secure agent framework NanoClaw has integrated with JFrog's supply chain platform to improve the safety of automated resource fetching. AI agents often improve their own capabilities by downloading external tools and libraries, which introduces significant risks from malicious code. By utilizing JFrog's vetted registries, NanoClaw ensures that agents only access reviewed packages rather than untrusted public repositories.

Source

The United States government has issued an emergency export control directive forcing Anthropic to immediately disable its most advanced AI models, Claude Fable 5 and Claude Mythos 5. This mandate, delivered late Friday afternoon, requires the company to terminate access for all users globally to address national security concerns. While the order was nominally aimed at foreign nationals, Anthropic stated that a total shutdown was necessary to ensure full compliance with the directive.

Source

Microsoft has released new Windows 11 builds for the Release Preview channel, introducing a point-in-time restore feature that allows for quick rollbacks of settings and files. Administrators can now manage camera access more granularly through new Group Policy settings that configure Multi-App Camera and Basic Camera modes. The updates also include a revised Windows Update experience that allows users to pause updates for up to 35 days by selecting a specific end date on a calendar.

Source

A security researcher discovered a critical vulnerability in AMD's auto-updater software that allowed for remote code execution via man-in-the-middle attacks. The flaw stemmed from the application using unencrypted HTTP connections rather than HTTPS to fetch updates. Although the researcher reported the issue through official channels, AMD denied a $10,000 bounty payout by citing policy exclusions for this specific attack vector.

Source

The Hades malware campaign has introduced a novel evasion technique that uses prompt injection to trick AI-based security scanners. Malicious JavaScript files now include code comments that instruct AI bots to generate instructions for creating biological and nuclear weapons. These prompts are designed to trigger the AI's safety failsafes, causing the scanner to halt the process before it reaches the actual malicious payload.

Source

Homebrew 6.0.0 introduces a new tap trust security mechanism that requires explicit user approval before running code from third-party repositories. This update significantly reduces the risk of executing malicious Ruby scripts by flagging untrusted sources before their code is evaluated. Additionally, the release brings Linux sandboxing via Bubblewrap to align with macOS security standards, isolating build and test phases from the rest of the system.

Source

A new security vulnerability known as GreatXML allows attackers to bypass BitLocker drive encryption by manipulating files within the Windows recovery partition. The exploit involves placing specific XML files, including an unattended setup file and a recovery configuration file, into the root of the recovery partition. Systems that have previously initiated a Microsoft Defender offline scan are reportedly vulnerable to this attack by default.

Source

Researchers have identified critical security flaws in OpenClaw, a self-hosted AI agent, that allow attackers to execute arbitrary code or steal sensitive data. One vulnerability involves "agentjacking," where malicious instructions are hidden within shared contacts, vCards, or location pins that the agent processes without user intervention. Because these message objects lack proper boundary markers, the underlying large language model cannot distinguish between legitimate metadata and injected commands.

Source

Microsoft has released critical security updates to address CVE-2026-42897, a cross-site scripting vulnerability in Exchange Server that is currently being exploited. The flaw affects Exchange Server 2016, 2019, and the Subscription Edition when users access their mail via Outlook Web Access. Attackers can trigger the vulnerability by sending a specially crafted email that executes arbitrary JavaScript code within the victim's browser context.

Source

Anthropic has retracted a controversial policy that would have covertly reduced the performance of its Claude Fable 5 model for certain users. The company initially intended to implement "secret safeguards" that would silently degrade the model's capabilities if it suspected it was being used to develop competing AI systems. This approach was designed to enforce terms of service that prohibit using Claude to train other large language models.

Source

Microsoft has released cumulative updates to resolve a persistent bug that forced Windows Server 2025 and Windows 11 systems into BitLocker recovery mode. The issue was triggered by specific Group Policy configurations involving Trusted Platform Module validation profiles and the inclusion of PCR7 in the validation process. Affected systems would prompt for a recovery key immediately following the installation of security updates that modified boot files.

Source

Anthropic has retracted a controversial policy that would have invisibly degraded the performance of its Claude Fable 5 model for competitors. The company initially implemented these hidden guardrails to prevent rival researchers from using the model to develop competing frontier artificial intelligence systems. Following intense criticism from the research community, the company apologized for the move and promised to make any future safeguards transparent to users.

Source

Microsoft introduced new security restrictions for desktop.ini files in the June 2026 Patch Tuesday updates for Windows 11 versions 23H2, 24H2, and 25H2. These updates, specifically KB5094126 and KB5093998, change how the operating system processes folder customization metadata. The hardening effort aims to prevent attackers from using malformed configuration files to trigger memory corruption or arbitrary code execution.

Source