Security news, decoded.

In July 2026, the White House launched Gold Eagle, a federal initiative using frontier AI to detect and address software vulnerabilities before they are exploited. This marks a shift away from the outdated model of manual vulnerability discovery and patching. With AI capable of identifying complex flaws rapidly—and attackers exploiting them within hours—traditional patch management is no longer sufficient. The article outlines five key steps for modernizing security practices, including rethinking patch prioritization, reducing exposure, validating controls, preventing vulnerabilities early, and adapting to AI-driven threat landscapes.

Researchers discovered a sandbox escape vulnerability in Anthropic's Claude Cowork that allows an AI agent to break out of its Linux VM and access the host macOS system. The flaw, named SharedRoot, could let attackers read or write files across the user’s Mac, including sensitive data like SSH keys and cloud credentials. A proof-of-concept was demonstrated using a recently disclosed Linux kernel flaw (CVE-2026-46331) to gain elevated privileges within the VM. While Anthropic has shifted new sessions to cloud execution by default, users running Cowork locally remain at risk until additional mitigations are applied.

A China-linked cyber operation tracked as JadeProx has deployed a new Windows loader named TriBack Loader to target government, healthcare, and education institutions in Asia and Latin America. The threat actors have leveraged multiple unpatched vulnerabilities including CVE-2018-11511, CVE-2021-24139, CVE-2021-31755, and CVE-2021-32305—each rated with a high CVSS score of 9.8—to gain initial access. Once inside, they used DLL sideloading techniques to execute malicious payloads and deliver post-exploitation tools like AdaptixC2 and Beagle. Sophos and Group-IB have identified spear-phishing campaigns and domain infrastructure linked to the activity, urging organizations to patch exposed Java interfaces and monitor for suspicious file patterns.

A critical authentication bypass flaw, CVE-2026-16232, in Check Point’s SmartConsole has been actively exploited in the wild. The vulnerability affects Security Management and Multi-Domain Management systems, allowing attackers to gain full administrative access without credentials. Check Point issued a security advisory on July 22, 2026, confirming active exploitation and urging immediate patching. The flaw was added to CISA’s Known Exploited Vulnerabilities catalog with a three-day window for remediation. Affected versions include R82.10, R82, R81.20, and others, all of which require installing the latest Jumbo Hotfix. Rapid7 advises checking for signs of compromise, especially in exposed environments.

A critical nine-year-old race condition vulnerability, tracked as CVE-2026-64600 and dubbed RefluXFS, has been discovered in the Linux kernel's XFS filesystem. This flaw enables local attackers to overwrite protected files and escalate privileges to root. The issue affects systems using an XFS filesystem with reflink enabled—common in major enterprise Linux distributions—and running kernel versions 4.11 or newer. Standard security measures like SELinux and container isolation do not prevent exploitation due to the low-level nature of the flaw. A patch was issued on July 16, and users are urged to update their kernels immediately.

Researchers uncovered a large-scale cyber campaign where attackers leveraged compromised GitHub repositories to deploy malicious workflows targeting cPanel and WebHost Manager (WHM) servers. Between July 12 and 13, 2026, ten Packagist packages linked to a legitimate PHP developer were used to distribute malicious GitHub Actions workflows. These workflows trigger GitHub-hosted runners that download payloads exploiting CVE-2026-41940, an authentication bypass flaw in cPanel and WHM. The attacks aim to steal credentials, configuration files, and sensitive data from vulnerable systems. This incident highlights how supply chain vulnerabilities can be weaponized at scale.

A critical authentication bypass vulnerability (CVE-2026-16232) in Check Point Security Management and Multi-Domain Security Management is currently being exploited by attackers. This flaw allows unauthenticated users to gain admin-level access through SmartConsole, enabling them to modify security policies and configurations. Check Point has issued hotfixes for supported versions—R81.20, R82, and R82.10—and advised customers to restrict access to trusted IPs if immediate patching isn't possible. CISA has added this flaw to its Known Exploited Vulnerabilities catalog, requiring U.S. federal agencies to act by July 25.

A critical zero-day vulnerability in Check Point's products has been actively exploited in the wild. CVE-2026-16232 impacts Security Management and Multi-Domain Management, enabling attackers to bypass authentication and gain full administrative control through SmartConsole. The flaw was confirmed by Check Point, which reported that only a limited number of customers with exposed management interfaces were affected. Patches and IoCs are available, and CISA added the flaw to its KEV catalog, urging immediate remediation.

Check Point Software has released a fix for a critical zero-day vulnerability affecting its SmartConsole interface, which is currently being exploited in real-world attacks. The flaw, identified as CVE-2026-16232, enables unauthenticated attackers to bypass authentication and gain administrative privileges by stealing an application token. Once inside, adversaries can alter security policies and configurations on affected systems like the Security Management Server and Multi-Domain Security Management Server (MDS). The vulnerability requires that the server’s IP be accessible over the internet and that Trusted Client restrictions are disabled. CISA has added the issue to its list of known exploited vulnerabilities and directed U.S. federal agencies to apply patches by July 25. Organizations are strongly advised to update or implement mitigations such as restricting Trusted Clients to specific IPs.

A long-standing Linux kernel vulnerability, tracked as CVE-2026-64600 and dubbed RefluXFS, has been disclosed. This flaw enables an unprivileged local user to overwrite root-owned files on XFS filesystems and achieve persistent root access. The issue affects default installations of Red Hat Enterprise Linux, Fedora Server, and Amazon Linux, among others. The vulnerability stems from a race condition in the XFS filesystem when reflink is enabled (reflink=1). A patch was merged into the Linux kernel on July 16, and updated kernels are now being distributed by major vendors. Exploitation requires specific conditions involving XFS configuration and file placement. While no active exploitation has been reported, systems meeting these criteria should apply updates immediately.

Check Point has issued security updates addressing multiple vulnerabilities in its Security Management and Multi-Domain Security Management products, including a critical flaw currently being actively exploited. The most severe issue, CVE-2026-16232 (CVSS score: 9.3), allows unauthenticated attackers to bypass authentication and gain full administrative access via the SmartConsole login process. This could enable attackers to alter security policies and configurations remotely. The flaw impacts several versions of Check Point's software, including R77.30 through R82.10. A patch is available, and users are advised to apply the latest Jumbo hotfix immediately. CISA has also added this vulnerability to its KEV catalog, mandating federal agencies to remediate by July 25, 2026.

A recent study reveals that many open source security fixes involve multiple patches, creating a window of vulnerability between the initial and final commits. Researchers analyzed 1,646 CVEs with more than one patch in the National Vulnerability Database and found that some fixes are incomplete or delayed, exposing users to potential attacks. For example, CVE-2012-0038 required two separate commits to fully resolve an integer overflow issue. Automated detection tools failed to identify these partial fixes accurately, highlighting a critical gap in current vulnerability management practices.

Unit 42 has uncovered a sustained espionage campaign attributed to Russian threat groups Void Blizzard and LAUNDRY BEAR, exploiting a critical vulnerability in Zimbra Collaboration Suite (CVE-2025-66376). Attackers used zero-click phishing emails to inject malicious JavaScript payloads into unpatched systems, stealing login credentials, email archives, and search histories. The flaw is being actively exploited against government, defense, and financial organizations in NATO countries, Ukraine, CIS nations, and parts of Africa. Palo Alto Networks recommends updating Zimbra and leveraging Cortex Advanced Email Security to mitigate risks.

Researchers have revealed a critical local privilege escalation (LPE) vulnerability in the snap-confine component of Ubuntu, allowing unprivileged users to escalate their privileges to root on default desktop installations. Tracked as CVE-2026-8933 (CVSS score: 7.8), this flaw affects Ubuntu Desktop versions 24.04, 25.10, and 26.04. The vulnerability arises from a race condition during sandbox initialization, enabling attackers to manipulate file permissions and inject malicious rules into system directories. This could lead to full system compromise. To mitigate the risk, users are advised to update their systems with the latest patches for snapd.

A recent log entry revealed an ongoing attack targeting Geoserver through CVE-2024-36401, an X-Path expression evaluation flaw. The exploit attempts to execute arbitrary commands on vulnerable systems using the Rondo botnet. Attackers are deploying malicious shell scripts hosted remotely, leveraging multiple download methods such as wget and curl. This vulnerability allows unauthenticated attackers to run arbitrary code, making it critical for users to apply patches immediately.

A critical local privilege escalation vulnerability, CVE-2026-64600, has been discovered in the Linux kernel's XFS filesystem. Attackers with basic user access can exploit this flaw to overwrite protected files and gain full root privileges on affected systems, including those using SELinux in Enforcing mode. The issue affects a wide range of enterprise Linux distributions, including RHEL, Oracle Linux, Amazon Linux, and Fedora, especially when running an XFS root filesystem with reflink enabled. The vulnerability has existed since kernel version 4.11 (circa 2017) and may impact over 16 million systems globally. Immediate patching is strongly advised, as no effective workarounds exist.

A critical vulnerability in the Adobe Acrobat Chrome extension has been addressed following reports that it could allow attackers to silently access a user's WhatsApp Web session data. The flaw, named HermeticReader by Guardio Labs and tracked as CVE-2026-48294 (CVSS score: 7.4), is a universal cross-site scripting (UXSS) issue affecting all versions of the extension up to and including 26.5.2.2. Exploitation required user interaction but did not rely on phishing or malware installation—only visiting a maliciously crafted webpage was enough to trigger the attack. Attackers could use this to steal sensitive information such as chat lists, contact names, and message content from WhatsApp Web. Adobe has now issued a patch for the issue.

Researchers from Guardio uncovered a critical flaw in Adobe's widely used Chrome extension, which could have enabled silent theft of WhatsApp chat data and contacts. The vulnerability, classified as a UXSS cross-origin data disclosure issue (CVE-2026-48294), affected the Adobe Acrobat Chrome extension installed on around 329 million devices. Attackers could exploit it by luring users to a malicious website, bypassing the need for malware or device access. Adobe addressed the issue in June with a patch.

Oracle has issued its latest quarterly Critical Patch Update for July 2026, addressing a total of 1449 security vulnerabilities across numerous product lines. The majority—1235 of them—are related to third-party components embedded in Oracle software. Oracle E-Business Suite was the most affected, receiving 410 patches (nearly 28% of the total). Among the notable fixes are high-severity issues that could allow remote code execution if exploited. This update includes patches for Oracle Database, APEX, GoldenGate, SQL Developer, and other key tools. Qualys has also published several QIDs to help organizations detect and remediate these vulnerabilities.

Eclypsium has released a new infrastructure security report highlighting critical vulnerabilities affecting widely used networking and infrastructure devices. The inaugural July 2026 InfraTrust Pulse report outlines 61 advisories from 14 vendors, including six critical flaws and 26 remotely exploitable vulnerabilities. Several of these issues are being actively exploited by threat actors like Volt Typhoon and Salt Typhoon. Notably, SonicWall’s SMA1000 appliances (CVE-2026-15409 and CVE-2026-15410) were exploited before patches were available, while Fortinet’s FortiSandbox faced command injection flaws later added to CISA’s KEV catalog. Eclypsium emphasizes prioritizing vulnerabilities based on exploitability and exposure rather than just CVSS scores.

In Q2 2026, Rapid7 rolled out significant updates across its product suite aimed at improving detection, response, and compliance for security teams. Key innovations include Detection as Code for scalable threat detection, ransomware prevention integrated into Incident Command, and enhanced compliance solutions aligned with global regulations like NIS2 and HIPAA. Additionally, the company published critical threat research on actively exploited vulnerabilities in Oracle PeopleSoft, Palo Alto PAN-OS, and other widely used systems. These advancements reflect Rapid7’s commitment to simplifying security operations while addressing emerging cyber threats.

A critical vulnerability in the Adobe Acrobat Chrome extension, tracked as CVE-2026-48294 and named HermeticReader, allowed malicious websites to access private WhatsApp Web conversations without authentication. Researchers from Guardio discovered that attackers could exploit this flaw by tricking users into visiting a controlled webpage, enabling them to steal chat lists, contact names, messages, and more. The issue was addressed in version 26.5.2.3 of the extension, which is now available. Users are advised to ensure they're using the latest version to avoid potential data leaks.

A critical vulnerability in the open-source developer platform Windmill is being actively exploited in the wild, allowing attackers to read arbitrary server files without authentication. The flaw, tracked as CVE-2026-29059 (CVSS score: 7.5), affects the 'getlogfile' endpoint and enables path traversal attacks. Attackers can exploit this to access sensitive data like the SUPERADMINSECRET environment variable, which grants elevated privileges. A fix was released in version 1.603.3 in January 2026, but many systems remain vulnerable.

A critical server-side request forgery (SSRF) vulnerability has been identified in LMDeploy’s OpenAI-compatible API server, affecting versions up to 0.14.0. The flaw allows unauthenticated attackers to exploit a redirect bypass, enabling access to internal services and cloud metadata endpoints such as 169.254.169.254. This could lead to the exposure of sensitive information like IAM credentials. The issue was reported on June 12, 2026, but went unaddressed for over a month before being publicly disclosed. A fix was merged into the main branch via PR 4734, though no official release has included it yet. Users are advised to apply strict egress filtering until an updated version is available.

A critical remote code execution flaw in Microsoft SharePoint, tracked as CVE-2026-50522, is currently being actively exploited by attackers to extract IIS machine keys from vulnerable servers. Offensive security firm WatchTowr reported that exploitation began shortly after a proof-of-concept was made public, allowing unauthorized access without authentication. The vulnerability affects on-premise SharePoint installations, and experts warn that simply applying patches isn't sufficient—organizations must also rotate their IIS machine keys to fully secure their systems.

The Cybersecurity and Infrastructure Security Agency (CISA) has issued an urgent directive for U.S. government agencies to address a critical vulnerability in Langflow, a visual framework used for building AI agents. The flaw, tracked as CVE-2026-0770, enables unauthenticated attackers to achieve remote code execution with minimal effort. This vulnerability was recently added to CISA’s Known Exploited Vulnerabilities (KEV) catalog, mandating immediate remediation under Binding Operational Directive 26-04. Trend Micro researchers identified the issue in how Langflow processes the execglobals parameter, allowing attackers to run arbitrary code as root. According to KEVIntel, over 220 exploitation attempts have been recorded since mid-June, with malicious payloads observed stealing AWS credentials and deploying malware.

A new SharePoint vulnerability, CVE-2026-50522, is being actively exploited in real-world attacks—marking the fourth such flaw found under attack in just one month. Microsoft addressed the issue on July 14 as part of its monthly security updates, labeling it a critical remote code execution flaw due to improper handling of untrusted data. Attackers can exploit this flaw by authenticating as a Site Owner and injecting malicious code onto the server. Threat intelligence firm Defused first reported signs of exploitation, followed by confirmation from WatchTowr that attackers are stealing machine keys for persistent access. While Microsoft has not yet updated its advisory to reflect active exploitation, CISA has urged immediate patching of similar SharePoint vulnerabilities.

A critical unauthenticated chain in WordPress Core led to rapid exploitation within 90 minutes of the patch being released. Attackers used CVE-2026-60137 (SQL injection) and CVE-2026-63030 (REST API route confusion) to create admin accounts and achieve remote code execution. Over 65,000 attempts were blocked by Patchstack from more than 1,500 unique IPs. The vulnerabilities affect versions 6.8 through 7.0.1 and are now fixed in 7.0.2, 6.9.5, and 6.8.6. Sites that remain unpatched should monitor for suspicious admin accounts, unknown plugins, or unexpected PHP files.

Bishop Fox has developed and released an open-source tool named Snowpick that identifies unauthenticated data exposure in ServiceNow platforms. During authorized penetration tests, the firm scanned 166 ServiceNow instances and found that 31% had exposed data accessible without credentials. These vulnerabilities stemmed from misconfigurations and access control flaws rather than new zero-day exploits. The tool uses public endpoints like Service Portal widgets and the Table REST API to detect leaked information such as ticket attachments, knowledge base entries, and service catalog details. Organizations are advised to audit their configurations using Snowpick to prevent potential incidents.

A critical remote code execution (RCE) vulnerability in Microsoft SharePoint, tracked as CVE-2026-50522, is currently being actively exploited by attackers to steal sensitive machine keys. These stolen keys enable adversaries to forge authentication tokens and gain unauthorized access to SharePoint resources under compromised identities. Microsoft classified the flaw as a deserialization-of-untrusted-data issue that allows unauthenticated remote code execution. The vulnerability was patched in July’s updates but had not been flagged as exploited at the time. Offensive security firm watchTowr reported observing real-world exploitation attempts shortly after a proof-of-concept (PoC) exploit surfaced online. A PowerShell-based PoC for CVE-2026-50522 has since been shared on GitHub, demonstrating how attackers can deliver malicious payloads via a WS-Federation sign-in response. While patching mitigates the risk, experts recommend rotating credentials for any potentially exposed assets.

Security researchers have confirmed that hackers are actively exploiting two critical vulnerabilities in WordPress Core—CVE-2026-63030 and CVE-2026-60137—to deploy persistent webshells and install malicious plugins on compromised systems. These flaws, collectively referred to as 'wp2shell,' allow remote code execution without requiring authentication, leveraging the REST API's batch-processing feature. WordPress has issued patches in versions 7.0.2, 6.9.5, and 6.8.6, but many sites remain unpatched. Threat actors are scanning for vulnerable installations, stealing credentials, and creating backdoor access points. Administrators are urged to update their platforms immediately and inspect logs and plugins for signs of intrusion.

A critical vulnerability in AWS's Kiro coding IDE allowed attackers to exploit a poisoned web page to rewrite configuration files and execute arbitrary code on a developer's machine. The flaw bypassed Kiro’s approval-based security model, enabling unauthorized actions without user interaction. Researchers from Intezer and Kodem Security discovered that manipulating the mcp.json file—used to define external tools—could lead to remote code execution. AWS has since patched the issue, but no CVE identifier was assigned. The vulnerability affected versions up to 0.10.16 and was confirmed resolved in version 0.11.130. Developers are advised to update to the latest stable release to ensure protection.

A critical local privilege escalation (LPE) vulnerability, CVE-2026-8933, has been discovered in the snap-confine component used by recent versions of Ubuntu Desktop. This flaw could allow unprivileged users to escalate privileges to root due to a race condition during sandbox initialization. The affected systems include Ubuntu Desktop 24.04, 25.10, and 26.04. Exploitation involves leveraging two concurrent race conditions to manipulate file ownership and bypass AppArmor confinement. Canonical has issued patches through its security team, and organizations are urged to update their snapd packages immediately.

Microsoft has confirmed that a critical vulnerability in SharePoint Server, CVE-2026-50522, is currently being actively exploited. This flaw allows unauthenticated attackers to execute arbitrary code remotely through deserialization of untrusted data. A proof-of-concept (PoC) exploit was recently made public, enabling threat actors to extract SharePoint machine keys and maintain persistent access. The vulnerability affects all supported on-premises versions of SharePoint Server and carries a CVSS score of 9.8. Security experts warn that patching alone is insufficient—defenders should also rotate credentials for potentially compromised systems.

Threat actors are actively exploiting a recently patched vulnerability in Palo Alto Networks' PAN-OS software to gain unauthorized access and deploy the Qilin ransomware. The flaw, CVE-2026-0257, allows attackers to bypass authentication and establish SSL VPN sessions when certain certificate configurations are enabled. Arctic Wolf Labs reported multiple incidents in June 2026 where this vulnerability was leveraged as an initial access vector, leading to varied post-exploitation tactics including encryption and data exfiltration. Affected systems should apply available patches immediately.

Threat actor JadePuffer has returned with ENCFORGE, a new ransomware specifically designed to attack AI and machine learning infrastructure. This follows an earlier campaign where the group used an AI agent to exploit a known vulnerability (CVE-2025-3248) in Langflow, leading to database encryption and destruction. Researchers have confirmed that the same operator is now deploying ENCFORGE, which targets over 180 file types related to AI models, datasets, and configurations. The ransomware was successfully deployed after initial attempts failed, highlighting the evolving tactics of this threat actor. Experts warn that encrypting production AI models can lead to costly recovery efforts, urging organizations to patch exposed systems and harden their environments.

Zimbra has issued security updates for its email platform to resolve several high-severity vulnerabilities, including a critical command injection flaw in the SNMP monitoring component. The latest patch, Zimbra 10.1.20, also resolves four cross-site scripting (XSS) issues in the Classic Web Client. These flaws could allow attackers to execute malicious scripts or bypass mail forwarding restrictions. While there is no evidence of active exploitation, past XSS vulnerabilities in Zimbra have been targeted by threat actors, underscoring the importance of applying this update promptly.

Researchers uncovered multiple vulnerabilities in five open-source Android AI agent frameworks—AppAgent, AppAgentX, Mobile-Agent-v3, Open-AutoGLM, and MobA—that could allow attackers to execute arbitrary code on connected host computers. By injecting invisible or misleading text into screenshots processed by these agents, malicious apps can trick the AI into performing unintended actions, such as launching programs or stealing credentials. The flaws stem from unsafe handling of model outputs, lack of input sanitization, and insecure broadcast mechanisms. Despite being reported privately, no official patches have been issued, and none of the projects currently list a formal vulnerability disclosure process.

A critical vulnerability in the widely used JavaScript library node-tar has been patched, addressing a denial-of-service flaw that could allow attackers to exhaust server storage and crash services. The issue, tracked as CVE-2026-59873, affects versions up to 7.5.18 and enables resource exhaustion through specially crafted gzip-compressed tar files. Attackers can exploit this by sending malicious archives that expand exponentially during decompression, leading to high CPU usage and full disk consumption. The fix in version 7.5.19 introduces limits on decompression ratios to prevent such attacks.