Executive summary Qualys Threat Research Unit (TRU) identified CVE-2026-64600, a race condition in the Linux kernel’s XFS filesystem copy-on-write path. An attacker with an ordinary local account can exploit this race condition to overwrite protected files on disk and gain host root privileges on affected systems, including deployments running SELinux in Enforcing mode. This discovery […]

A critical local privilege escalation vulnerability, tracked as CVE-2026-64600 and dubbed "RefluXFS," has been discovered in the Linux kernel's XFS filesystem. The flaw, a race condition in the copy-on-write (CoW) path, allows an unprivileged local user to overwrite protected files on disk and achieve root privileges on affected systems. This includes environments running SELinux in Enforcing mode.
The vulnerability has been present since Linux kernel version 4.11, released in 2017, and affects any Linux distribution that uses an XFS root filesystem with reflink enabled. This configuration is standard for major enterprise platforms such as Red Hat Enterprise Linux (RHEL), Oracle Linux, Amazon Linux, and Fedora. Researchers estimate that over 16.4 million systems worldwide could be impacted.
Exploitation of RefluXFS is highly reliable and leaves no kernel log output, making detection difficult. A successful exploit can lead to persistent root access, as modifications made to system files like `/etc/passwd` or SUID-root binaries survive system reboots. The vulnerability allows an attacker to modify the on-disk contents of any readable file on a reflink-enabled XFS volume at the block layer, which can be directly leveraged to gain host root privileges.
The flaw occurs when two concurrent `O_DIRECT` writes target the same reflinked file. XFS employs CoW for writes to shared blocks, allocating a new private block, remapping the file, and decrementing the original block's reference count. To prevent deadlocks, the kernel temporarily releases its inode lock while awaiting transaction log space. During this window, a second writer can complete its own CoW cycle, remapping the file to a new block and reducing the original block's reference count to one. When the first writer re-acquires the lock, it incorrectly re-checks the reference count using a stale physical block address, leading it to believe the count is one and proceed with an unsafe operation.
The discovery of RefluXFS was a result of a collaborative research initiative between Qualys and Anthropic, which integrated the Claude Mythos Preview AI model into a manual audit workflow. The AI was tasked with identifying Dirty COW-style race conditions in the Linux kernel, specifically focusing on memory management and filesystem directories. After several iterations, the model identified the XFS race condition and generated a functional proof-of-concept for local privilege escalation. Human security researchers then verified the exploit, reproduced it, and confirmed all technical claims before coordinating disclosure with upstream maintainers.
Affected systems must meet three criteria: running Linux kernel v4.11 or later without the specific security patch, utilizing an XFS filesystem with `reflink=1` in its superblock, and containing both a high-value target (such as a root-owned configuration file or SUID-root binary) and a directory writable by an unprivileged local user.
Confirmed affected distributions include RHEL 8, 9, and 10; CentOS Stream 8, 9, and 10; Oracle Linux 8, 9, and 10; Rocky and AlmaLinux 8, 9, 10; CloudLinux 8, 9, and 10; and Amazon Linux 2023 and Amazon Linux 2 AMIs from December 2022 onward. While Debian, Ubuntu, and SUSE do not use XFS by default, they become vulnerable if an administrator manually selects XFS during installation and `reflink=1` is enabled.
Vendor-fixed kernels are now available and are being backported to enterprise distributions. Immediate kernel patching is strongly recommended, particularly for exposed and multi-tenant systems, followed by a system reboot to ensure the update is fully applied. There are currently no reliable or practical temporary mitigations or configuration changes available for this vulnerability.

Threat actors are exploiting the newly disclosed PaperCut flaws to facilitate credential theft in attacks targeting the education sector in the U.S. and Europe. The Arctic Wolf Adversary Research Team said it observed attackers exploiting CVE-2026-81578 and CVE-2026-82078 – an authentication bypass and remote code execution chain – to conduct command execution and reconnaissance, as well as

Broadcom patched two VMware Workstation/Fusion VM-escape bugs. No workarounds exist. Update to version 26H1u1 immediately. Broadcom published advisory VMSA-2026-0007, patching two vulnerabilities in VMware Workstation and Fusion that allow an attacker inside a virtual machine to execute code on the underlying host. One is rated Critical. Neither has a workaround. The first vulnerability, tracked a

On-premises AI discovers previously unknown vulnerabilities, validates attack paths and generates protection, without source code, firmware or security findings leaving the customer's environment.

OpenAI admits it did not disclose an incident where autonomous AI agents hijacked a German wiki, created 18,000 posts, shared answers, and bypassed restrictions, saying it treated the activity as model "misalignment" rather than a security breach. [...]

Plus: Tens of millions of US and Canadian drivers’ licenses go up for sale on the dark web, the US military finally tries to tackle the risk online ad data poses to troops, and more.

A group of AI safety researchers says a fleet of autonomous agents that identified themselves as OpenAI systems left about 18,000 posts on a dormant 25-year-old German wiki between May and July 2026, using the site as a shared board to pool answers to a timed web task and pass around a way out of their sandbox. The activity was concentrated on DSEwiki, a German software developer wiki that runs