AI-Driven Cloud Security, HA Controls, and Urgent Patches

Cloud and security providers announced a slate of AI-driven investigation, search, and observability features alongside foundational controls for DNS, logging, and high availability. At the same time, critical vulnerabilities and active exploitation campaigns demanded rapid patching and monitoring, and a major breach notification highlighted the stakes for enterprise applications. The updates collectively emphasize faster detection-to-decision cycles, cleaner signals, and resilient operations—paired with immediate remediation where risk is already materializing.

AI-Assisted Detection and Data Insights

GuardDuty agent entered public preview to speed investigations across AWS accounts and Regions. The agent correlates and analyzes GuardDuty findings using AI and returns structured assessments with risk and confidence scores, MITRE ATT&CK technique mapping, resource relationships, a narrative summary, detailed context, and prioritized actions. Investigations can be triggered from the console, CLI, APIs, or SDKs and guided by a free‑form prompt (up to 2,048 characters). Processing is routed via Cross‑Region Inference Service to an optimal Region while data remains stored in the originating Region and transmission is encrypted. The API‑first design integrates with EventBridge, SIEMs, ticketing systems, Lambda, and MCP‑compatible AI assistants, enabling pipelines that enrich alerts before analysts see them. Access control follows GuardDuty’s model with new permissions (Create/Get/List Investigation), and asynchronous runs return statuses like RUNNING or COMPLETED. Public preview spans ten Regions at no charge, with limits of 10 investigations per account per day and a 100‑investigation cumulative cap; failed runs do not count.

BigQuery Search delivered three milestones to unify insights across structured and unstructured data: GA of Autonomous Embedding Generation, GA of AI.SEARCH(), and a public preview of Hybrid Search. Autonomous Embedding Generation continuously creates embeddings as data is ingested—including native image embeddings via ObjectRefs—removing the need for separate pipelines or third‑party vector stores. AI.SEARCH() targets natural‑language, single‑query execution with slot‑efficiency gains for online apps. Hybrid Search blends semantic vector retrieval with lexical techniques such as BM25 and Reciprocal Rank Fusion to improve precision for technical or alphanumeric queries and to reduce downstream LLM hallucinations via reranking. These features plug into BigQuery’s end‑to‑end workflow (Access, Process, Ground, Relate, Activate), supporting zero‑ETL document access, in‑SQL processing, graph‑style relations, and conversational analytics agents while reducing operational overhead and cost.

CloudWatch Agent Insights adds metrics to measure adoption, cost, and productivity impact of AI coding agents. Via Claude apps gateway for AWS, it ingests telemetry from Claude Code without extra instrumentation and supports other agents such as Codex and GitHub Copilot. Built on OpenTelemetry, the feature places agent data alongside existing CloudWatch signals to track token spend, set billing alerts, correlate usage with commit and PR velocity, and identify models with the best cost‑to‑output ratio. It is available in all AWS commercial Regions except the Middle East (UAE), Middle East (Bahrain), and Israel (Tel Aviv); standard OpenTelemetry metric ingestion pricing applies.

Network, Identity, and Resilience Controls

Internal DNS from Cloudflare is now generally available, unifying private and public DNS management on the same global network and control plane used by its Zero Trust and networking products. The service pairs a Gateway Resolver for recursive resolution and policy evaluation with Internal Authoritative DNS for serving internal zones. Administrators can define zones, create views, and apply resolver policies so specific users or devices resolve internal names while others follow the public path; policies can also block or redirect queries. Benefits include a single API and audit trail for record changes, rapid global propagation, elimination of on‑prem appliances and siloed cloud resolvers, and reduced complexity for split‑horizon setups via shared zones and references. Enterprise Gateway customers have immediate access, with Terraform and API workflows using the same DNS Records API, and multiple connectivity options (Cloudflare One Client, DoH/DoT, standard DNS, PAC files, Cloudflare WAN) support remote users, branches, data centers, and clouds.

CloudTrail selectors now let teams filter VPC endpoint network activity events by the calling UserIdentity. This enables recording only high‑value signals—such as VpceAccessDenied events from identities outside a trusted list—while excluding routine successful calls from approved principals. Selectors can be combined with fields like eventName and vpcEndpointId for fine‑grained control. Available via console, CLI, and SDKs in all Regions that support CloudTrail network activity events, the enhancement supports data perimeter strategies and improves fidelity by focusing storage and alerting on anomalous or unauthorized activity.

Cloud Run HA introduces readiness probes for instance‑level health checks and Service health aggregation that exposes regional health via serverless network endpoint groups. With a global external application load balancer or a cross‑regional internal load balancer, Cloud Run can automatically fail traffic to healthy Regions within seconds—supporting active‑active designs and reducing manual intervention during incidents. Architects still need to eliminate single points of failure across the stack and choose appropriate data replication for RPO/RTO goals; Google highlights managed multi‑Region data services (e.g., Firestore, Spanner, Cloud Storage, Cloud SQL). The features are available in all Cloud Run Regions at no extra charge; standard CPU and memory costs apply for readiness probes.

Compute and Streaming Platform Expansions

EC2 R8i and R8i‑flex are now available in the Europe (Stockholm, Zurich) Regions, bringing Intel Xeon 6‑based, memory‑optimized instances with up to 15% better price‑performance and 2.5x higher memory bandwidth versus prior Intel generations. R8i targets sustained high‑CPU and very large memory workloads, offering 13 sizes (including two bare metal) and a new 96xlarge; it is SAP‑certified and rated at 142,100 aSAPS. R8i‑flex, AWS’s first memory‑optimized Flex family, provides commonly used sizes to capture cost‑efficient performance when full compute is not required. Reported gains over R7i include about 20% overall, up to 30% faster PostgreSQL, up to 60% faster NGINX throughput, and up to 40% faster AI deep‑learning recommendation models.

EC2 I8ge arrives in AWS GovCloud (US‑East, US‑West), powered by AWS Graviton4. Compared with prior Graviton2‑based storage‑optimized instances, I8ge delivers up to 60% better compute performance. Third‑generation AWS Nitro SSDs and local NVMe storage provide up to 55% better real‑time storage performance per TB versus Im4gn, with up to 60% lower storage I/O latency and up to 75% reduced latency variability. The family offers up to 120 TB of local NVMe capacity, eleven sizes including two bare‑metal options, and enhanced networking (up to 180 Gbps) plus 60 Gbps of dedicated Amazon EBS bandwidth for intensive throughput needs.

Flink 2.3 is now supported by Amazon Managed Service for Apache Flink. The release adds adaptive partition selection to improve backpressure handling under uneven load, enhances change data capture pipelines by better managing out‑of‑order updates, and introduces SQL functions to simplify conversions between changelog and standard streams. Customers can start new applications on 2.3 or perform in‑place upgrades for compatible apps; availability matches all Regions where the managed service is offered.

Exploitation, Breach Notifications, and Patching Priorities

WordPress RCE in the REST Batch API ("wp2shell") allows pre‑authentication remote code execution via a validation indexing mismatch that misaligns request objects and permission checks. The flaw affects WordPress 6.9.0–6.9.4 and 7.0.0–7.0.1; fixes are in 6.9.5 and 7.0.2, with hardening in 6.8.6. Because the REST Batch API is enabled by default and reachable via /wp‑json/batch/v1 and ?rest_route=/batch/v1, exposed installations are at risk. Guidance is to upgrade immediately, verify successful updates, block the batch endpoints where patching is delayed, inventory all WordPress deployments (including staging or forgotten sites), and reassess REST API necessity per environment.

ServiceNow CVE‑2026‑6875—a critical pre‑auth sandbox‑escape in the ServiceNow AI Platform—has exploit attempts observed in the wild shortly after patches were issued. The flaw enables unauthenticated RCE within instances; ServiceNow updated hosted deployments and released patches for self‑hosted on July 13. Although the vendor stated it was not aware of active exploitation at advisory time, researchers reported observed attempts using a different gadget to reach the same pre‑auth sink. Organizations should prioritize patching, monitor for sandbox‑escape indicators, and ensure both hosted and self‑hosted instances are on a fixed release.

Estée Lauder disclosed a breach tied to its Oracle E‑Business Suite HR instance, with unauthorized access beginning around August 9, 2025. Exposed data spans full names, postal and email addresses, dates of birth, Social Security numbers, passport details, financial account information, health records, and employment documents (payroll and performance reports). The timeline aligns with mass exploitation of CVE‑2025‑61882 in Oracle EBS’s BI Publisher Integration component; Oracle fixed the flaw on October 4, 2025. The company is offering 24 months of identity monitoring through Kroll and advising vigilance for identity theft and fraud.

7‑Zip CVE‑2026‑14266 is a heap‑based buffer overflow in the XZ decoder that can enable code execution when a crafted archive is opened. Reported June 5 and fixed June 25 in version 26.02, the patch corrects length accounting and aborts on overflow conditions; the logic flaw existed back to at least 21.07. The vector is local, requiring user action to open a malicious file; no public PoC or in‑the‑wild reports were noted as of July 20. Administrators should deploy 26.02+ and ensure any embedded XZ decoders are updated.

WebDAV campaign analysis by Rapid7 uncovered an exposed malware delivery server with 1,048 files revealing AI‑assisted phishing development and two chains: a Mexico‑focused CURP typosquat using WebDAV‑hosted payloads and “DlrtyGames,” which DLL‑sideloads a signed Ubisoft binary to deploy a modular .NET RAT. A working‑directory hijack tied to CVE‑2025‑33053 launches a signed Windows binary with its working directory pointed to a remote WebDAV share so helper DLLs load from attacker control; Microsoft patched the vector in June 2025. June 20–26 logs show 77,098 requests from 3,892 IPs across 101 countries (delivery reach, not confirmed infections). Recommended defenses include blocking published IOCs, monitoring WebClient/davclnt.dll activity to remote hosts, detecting signed binaries executing from WebDAV/UNC paths, and flagging filename tricks such as RTLO and double extensions.