Patches Intensify as EU Pressures Google; Cloud Vendors Add Controls

Emergency fixes for widely deployed software, new additions to CISA’s Known Exploited Vulnerabilities catalog, and a landmark European Union mandate on Android assistants set the tone for the day. Cloud providers also introduced governance and AI platform updates aimed at tightening controls and improving operational resilience. Several organizations reported or settled incidents, underscoring persistent exposure from third-party systems, credential abuse, and novel persistence techniques.

Urgent Patches: Web Stacks Under Pressure

A critical NGINX vulnerability (CVE-2026-42533) was patched in nginx 1.30.4 (stable), 1.31.3 (mainline), and NGINX Plus 37.0.3.1. The heap buffer overflow arises from a specific configuration pattern that combines regex-based maps and subsequent string expressions referencing earlier regex captures. Because nginx evaluates in two passes, a regex run between passes can corrupt capture state, causing worker crashes and potential remote code execution where ASLR is disabled or bypassable. F5 rated the issue highly (CVSS v4 9.2; v3.1 8.1) and urged immediate upgrades. A temporary mitigation—converting affected regex maps to named captures—may not cover all variants; upgrading to patched releases remains the only comprehensive fix.

Administrators of WordPress sites face active exploitation risk from the pre-authentication RCE chain dubbed WordPress RCE “wp2shell,” which links a REST API batch-route confusion bug (CVE-2026-63030) with an SQL injection in WP_Query (CVE-2026-60137). Affected ranges are 6.9.0–6.9.4 and 7.0.0–7.0.1; patched versions 6.9.5 and 7.0.2 were released with forced automatic security updates. Public proof-of-concept exploits and signs of in-the-wild activity increase urgency. Short-term mitigations include blocking anonymous REST API access or the /wp-json/batch/v1 and ?rest_route=/batch/v1 endpoints at the WAF, but patching is required for full remediation and defenders should monitor for indicators such as unexpected admin logins, new plugins, or remote command execution artifacts.

KEV Mandates and Industrial Exposure

CISA issued a CISA mandate for federal agencies to urgently patch two actively exploited Fortinet FortiSandbox command injection vulnerabilities (CVE-2026-39808 and CVE-2026-25089), both rated 9.1 CVSS. Fixes are available in FortiSandbox 4.4.9 and 5.0.6, with exposure spanning multiple 4.4.x, 5.0.x, and 4.2 versions as well as specific cloud and PaaS builds. CISA added both to the KEV catalog and set a July 19 remediation deadline, advising discontinuation of cloud-based services if mitigations are not possible. Organizations running affected versions should prioritize patching, validate deployments, and implement compensating controls where needed.

Alongside Fortinet, a critical SharePoint Server flaw, SharePoint RCE CVE-2026-58644 (CVSS 9.8), was added to KEV after confirmation of exploitation before July Patch Tuesday fixes. The vulnerability is a deserialization of untrusted data that allows an attacker authenticated at least as a Site Owner to inject and execute arbitrary code remotely. CISA urged immediate patching and hardening, including enabling AMSI integration, removing intrusion artifacts before rotating IIS keys, implementing tailored logging to detect exploitation, and avoiding direct internet exposure of SharePoint servers and administrative interfaces.

In operational technology, Unit 42 and Siemens detailed a chained zero-day trilogy affecting Ruggedcom ROX II, disclosed as Unit 42 research. The issues—CVE-2025-40948 (arbitrary file disclosure), CVE-2025-40947 (command injection/privilege escalation), and CVE-2025-40949 (persistent root command execution)—enable progression from reconnaissance to full root control and persistence. Siemens released advisories (SSA-973901, SSA-078743, SSA-081142) and recommends updating to firmware V2.17.1. Unit 42 notes options for virtual patching and detection via Palo Alto Networks controls. The findings highlight risks in devices used for network segmentation and stress rigorous input validation and secure privileged command handling in critical industrial environments.

Cloud Governance and AI Platforms

Google Cloud introduced a preview of IAM-based BigQuery tags for scalable column-level security. Defined at the organization level as purpose=DATA_GOVERNANCE, these global tags support hierarchical trees up to five levels and automatically replicate to secondary regions. A three-step workflow covers creating tag keys/values, attaching tags to table columns via schema updates or SQL DDL with verification in INFORMATION_SCHEMA, and creating regional data policies to grant raw access or apply masking (e.g., SHA256, ALWAYS_NULL). Effective protection remains layered—users still require base table access—and tags are global while data policies are regional, enabling centralized governance and resilient multi-region operations.

For teams building AI agents, Google published 13 modular Gemini demos on the Enterprise Agent Platform using the ADK and Agents CLI. Tutorials span ADK fundamentals, human-in-the-loop approvals, MCP tools for data access, and Agent-to-UI patterns for dynamic frontends. Runtime topics include stateful data science agents, durable workflows that survive restarts, and deployment of an ambient expense agent with integrated tracing and analytics. Governance and quality features include secure coding practices, an Agent Gateway for mTLS/IAM-protected outbound calls, Model Armor for content inspection, and evaluation flywheels powered by adaptive AutoRaters, with orchestration examples across A2A, LangGraph, and CrewAI.

Amazon SageMaker HyperPod added partition-level topology for Slurm clusters in an HyperPod update, enabling different network topologies to coexist within a single cluster. UltraServer instance families (e.g., ml.p6e-gb200.36xlarge) map to block topology, while hierarchical-interconnect families (e.g., ml.p5.48xlarge, ml.p5e.48xlarge, ml.p5en.48xlarge) map to tree topology. HyperPod maintains per-partition assignments through scaling and node replacement, improving GPU-to-GPU communication and NCCL efficiency. Topology-aware scheduling is on by default for Slurm 25.11+ with no extra configuration, and availability spans all Regions where HyperPod is supported.

Policy Shifts and Active Intrusions

The European Commission issued binding DMA specifications requiring Google to open Android capabilities to rival assistants and to share anonymized search data, as detailed in this EU order. Eleven OS features are covered: five “restricted” functions—available to certified assistants via a Qualified AI Assistant Programme—include centralized on-device app data, context-aware intelligence, structured app integration, screen automation, and deep system integration; six are opened without certification (e.g., ambient sensor data, hotword detection, background execution). The search data directive includes a three-step anonymization process and strict contractual controls. Deadlines include draft terms by February 1, 2027, final by May 1, 2027, and implementation by Android 18 or August 1, 2027 (hotword concurrency by Android 19/August 1, 2028). Google criticized the measures over privacy and security concerns, while the Commission set oversight requirements including published terms and acceptance of third-party Trusted Certification Authorities.

A bipartisan coalition of 42 U.S. attorneys general reached an $18 million 23andMe settlement tied to a 2023 credential-stuffing incident that exposed profile and ancestry data for more than six million customers. The resolution, linked to the firm’s bankruptcy proceedings and data transfer to TTAM Research, mandates structural safeguards: risk analysis, a data security Advisory Board at TTAM, and preserved consumer deletion rights. International regulators previously fined 23andMe in the UK and Spain over the incident, and a separate victim compensation settlement was approved by a bankruptcy judge.

Ernst & Young disclosed a third-party support ticket compromise in an EY breach that saw an unauthorized actor access and download multiple documents between March 28 and April 12. The files may include personal and financial data used for tax filings. EY removed the access, notified law enforcement, and is offering 24 months of Experian identity monitoring to affected clients.

On Windows, a newly published proof-of-concept called the LegacyHive exploit targets an unassigned CVE in the User Profile Service to escalate privileges by loading and modifying a target user’s registry hive. Independent testing confirmed that successful exploitation lets non-admin users modify the classes hive and achieve automatic code execution when an administrator signs in. Detection guidance and Defender queries were quickly shared; Microsoft had not provided immediate comment to the outlet.

Kaspersky documented a stealthy OAuth-based mailbox persistence technique, “Shadow Token via Remote Debug (STRD),” in a Shadow Token report. Attackers duplicate Chrome/Edge profiles, run a browser in debugging mode, automate OAuth consent via DevTools, and obtain tokens hosted on attacker infrastructure to access Google Workspace mailboxes—potentially surviving password resets. Recommended defenses include auditing and revoking unnecessary OAuth app permissions, blocking non-developer browser debugging with policy, and monitoring for browser launches with active debugging ports.