GovCloud AI Expands, Agentic Governance Advances, and New…

Public cloud providers rolled out regional AI and analytics capabilities while publishing guidance on tighter governance for agentic systems. Alongside these updates, researchers detailed high-impact vulnerabilities affecting virtualization, enterprise software, and AI platforms, and a major services firm confirmed a code-related breach claim. The day’s developments underscore a dual focus: expanding performance and compliance-ready options while reducing exposure through architectural controls and timely patching.

GovCloud AI Data and Analytics Options Expand

Amazon Web Services extended S3 Vectors to the AWS GovCloud (US-East and US-West) Regions. The managed vector store brings dedicated APIs and S3-aligned durability and availability to workloads such as agents, inference, Retrieval Augmented Generation, and semantic search at scale, allowing public-sector and regulated customers to deploy vector-backed applications within compliance-focused boundaries.

Amazon also launched Redshift RG instances in GovCloud (US). Powered by AWS Graviton and a vectorized data lake query engine, RG targets large analytics and data lake scenarios with up to 2.4x performance gains over RA3 and lower price per vCPU. Customers can migrate via Snapshot & Restore or resize paths and choose among On-Demand and Reserved Instance options.

Regional Performance and Cost Changes for Compute and Storage

EC2 C8ine instances are now available in Europe (Frankfurt), built on custom sixth-generation Intel Xeon Scalable processors and sixth-generation AWS Nitro accelerators. AWS reports up to 43% higher compute performance versus C6in, up to 2.5x higher packet performance per vCPU, and up to 2x greater Internet gateway throughput, benefiting virtual firewalls, load balancers, and telco network functions.

S3 Express One Zone has expanded to the Europe (Frankfurt) Region. The single-AZ, high-performance storage class targets latency-sensitive workloads—ML training, interactive analytics, AI search key-value caching—with claims of up to 10x faster access and up to 80% lower request costs compared to S3 Standard, while maintaining high durability and availability within one Availability Zone.

GPU management fees were reduced for managed container and Kubernetes environments. For Amazon ECS Managed Instances, ECS Managed fees drop by 35% for G-series and 60% for P-series and AWS Trainium instances; identical reductions apply to EKS Auto Mode. Changes are automatic for existing clusters and are available across supported Regions.

EMR Serverless added larger worker sizes—up to 32 vCPUs and 244 GB memory—doubling previous limits. The enhancement aims to reduce shuffle overhead, mitigate out-of-memory failures from skewed datasets, and enlarge in-memory caches for Spark and Hive workloads, potentially improving runtimes and cost profiles.

Governing Agentic AI and Advancing In‑Database Capabilities

Google Cloud published an overview of its State of AI Infrastructure, noting that 83% of organizations need upgrades to support agentic AI. The Google Report highlights demand for fluid compute (e.g., TPU variants and Arm-based CPUs), governance and security cited by 79% of leaders, widespread hybrid multicloud and edge importance, and energy constraints shaping hardware choices. A companion guide details how to build and list agents as services in Gemini Enterprise and Marketplace, covering the A2A protocol, OAuth Dynamic Client Registration, procurement flows, and governance roles; see Gemini Agents for implementation steps.

AWS outlined how to enforce zero data retention for Amazon Bedrock at multiple control layers. The Bedrock Retention guidance explains retention modes (none, default, inherit, provider_data_share), differences between bedrock-mantle and bedrock-runtime endpoints, and the use of AWS Organizations SCPs to prevent enabling provider data sharing. It also notes operational impacts where specific APIs require retention and would be blocked under a zero-retention setting.

Amazon RDS for Oracle now supports Oracle Database 26ai. With RDS for Oracle, organizations can use Select AI for natural-language SQL generation and Oracle AI Vector Search for RAG directly from SQL, alongside JSON Relational Duality Views and SQL Property Graphs. The release is Enterprise Edition only and available in commercial and GovCloud Regions, with upgrade and migration paths provided.

Cloudflare joined the UK government’s Cyber Resilience Pledge as a founding signatory. The Cloudflare Pledge post emphasizes leadership accountability, transparency, and baseline controls, and outlines contributions such as unmetered DDoS mitigation, DNSSEC support, post-quantum crypto deployments, and governance practices.

Exploited and Emerging Vulnerabilities, Plus One Corporate Breach

Researchers disclosed a long-standing Linux kernel KVM/x86 use-after-free vulnerability, CVE-2026-53359 (“Januscape”), enabling guest-to-host escape on Intel and AMD systems. The Januscape analysis notes a patch landed in June; exploitation from a VM with root can crash the host or execute code on it, threatening co-resident VMs, with elevated risk where /dev/kvm permissions are overly broad.

Adobe issued emergency fixes for multiple ColdFusion flaws, with CVE-2026-48282 (path traversal leading to code execution) targeted within hours of disclosure. Administrators are urged to patch per APSB26-68, as reported in Adobe ColdFusion, and to harden exposed instances amid observed exploitation and a faster advisory cadence.

BeyondTrust warned of critical vulnerabilities in Remote Support and Privileged Remote Access that could allow unauthenticated access under specific configurations. Patches were applied to cloud-hosted instances in April; self-hosted customers should update to the April rollup or RS/PRA 25.3.3 or later, per the BeyondTrust Advisory.

Noma Security showed how a single public GitHub issue could manipulate previewed agents to exfiltrate private repository data via indirect prompt injection against GitHub Agentic Workflows. The GitHub Workflows report recommends scoping tokens narrowly, restricting triggers, limiting outputs, and adding human review, noting that filtering alone is insufficient where agent credentials and untrusted inputs converge.

Mandiant described recovering active ADFS token-signing keys from the machine-scoped DPAPI store when configuration drift leaves a “ghost” key in the ADFS database. The Mandiant Post details detection via auditing of MachineKeys/DPAPI directories and recommends treating ADFS as Tier 0, migrating keys to HSMs, and ensuring certificate rotations update ADFS configuration.

Sand Security disclosed a session isolation flaw in the Writer platform that could exfiltrate session tokens across tenants via agent preview links. Following disclosure, the vendor isolated cookies to a separate origin and prevented forwarding into previews, blocking the attack path; see Writer Flaw for details on root cause and remediation.

Accenture confirmed an isolated security incident after a threat actor advertised data allegedly taken from an Azure DevOps repository, offering roughly 35 GB of source code and artifacts for sale. The company reported remediation and no impact to operations; further details remain limited, as covered in Accenture Breach.