Cloud Identity, Routing Security, and Critical Patch Priorities

Cloud providers shipped notable identity, networking, and AI platform updates alongside new research spotlighting risks in webmail and NAT state handling. Several critical fixes and active exploits require prompt action across BI platforms, load balancers, CMS, and RMM software. Breach disclosures and an ongoing vishing-driven extortion campaign further underscored the importance of phishing-resistant authentication, supply chain hygiene, and post-incident reviews.

Raising the Bar for Cloud Identity, Governance, and AI Operations

AWS IAM Identity Center introduced a one-click multi-Region option for new organization instances, automating creation of a customer managed multi-Region KMS key and replication to an additional supported Region. The update aims to simplify deployment of resilient workforce access, with administrators still able to choose single-Region or custom configurations (including bringing an existing KMS key). The capability is available in 17 commercial Regions enabled by default for organization instances; standard AWS KMS charges apply, and IAM Identity Center itself adds no extra cost.

Amazon added Amazon Cognito as a core skill to the Agent Toolkit for AWS, enabling AI coding agents to automate secure sign-in workflows end to end. The new Cognito skill covers user pools and app clients, OAuth 2.0 flows, token handling and JWT authorizers, passkey/WebAuthn enrollment, threat protection, Lambda triggers, and identity pools. Agents can execute AWS CLI commands directly or work through the AWS MCP Server to enforce IAM-based guardrails, with actions recorded in CloudTrail for auditability.

Amazon Bedrock expanded into AWS GovCloud (US-West) with new AgentCore capabilities. Bedrock AgentCore now provides short- and long-term memory, a policy subsystem that converts natural-language policies into Cedar for gateway enforcement, and a managed harness that lets teams declare models, tools, and instructions in configuration. The combination is designed to accelerate governed, context-aware agent deployments in regulated environments while centralizing controls over tool access and observability.

AWS also expanded compliance coverage for its HPC offering: AWS PCS is now in scope for FedRAMP Class C in several commercial US Regions and Class D in AWS GovCloud (US), as well as SOC, ISO, CSA STAR, and PCI attestations. PCS is also HIPAA eligible. These attestations support deployment of sensitive or mission-critical HPC workloads while meeting governance and audit requirements, with customers advised to confirm region-specific scope before use.

Cloudflare is unifying its AI Gateway and Workers AI into a single control plane. The Cloudflare AI control plane exposes a unified /ai/ API and AI binding, auto-creates a default gateway on first use for request/response logging and token accounting, and enables unified billing so Gateway credits can be used with Workers AI. Cloudflare is shifting from provider-first to model-first routing with planned intelligent selection, failover, and load balancing across vetted providers; users can benefit immediately from logging and simplified integration via the default gateway.

AWS broadened compute options for latency- and memory-sensitive workloads. EC2 R8i and R8i-flex instances are now available in Europe (Milan), powered by custom Intel Xeon 6 processors exclusive to AWS. AWS reports up to 15% better price-performance and 2.5x higher memory bandwidth versus prior Intel-based generations, with gains up to 60% for NGINX web apps, 30% for PostgreSQL, and 40% for certain deep learning recommendation models. R8i-flex debuts as the first memory-optimized Flex family for workloads that do not require full compute capacity. The R8i family spans 13 sizes (including two bare metal options and a new 96xlarge) and is SAP-certified at 142,100 aSAPS.

For game backends, Amazon GameLift Servers added 21 new EC2 instance types across compute-optimized and general-purpose families. GameLift Servers now supports C8a, C8i, C9g and M8a, M8i, M9g, offering a mix of 5th-gen AMD EPYC, custom Intel Xeon 6, and AWS Graviton5. The expansion improves flexibility for CPU-intensive multiplayer logic and cost-optimized fleets; availability follows GameLift Servers’ supported Regions, excluding AWS China.

To ease maintenance for search and analytics clusters, AWS extended its upgrade runway and published support timelines for additional engine versions. OpenSearch Extended Support now continues security and OS patches for specific Elasticsearch 1.5–7.8, OpenSearch 1.0–1.2, and OpenSearch 2.3–2.9 versions through November 7, 2027. From November 7, 2026, a surcharge equal to instance pricing applies for those versions (storage pricing unchanged), with example surcharges listed at $0.0065 per Normalized Instance Hour in us-east-1. AWS also detailed timelines for Elasticsearch 6.8, 7.9, 7.10 and OpenSearch 1.3, 2.11–2.19, and recommends upgrading to current releases for performance and security enhancements.

Routing Hygiene and NAT-State Risks Move Into Focus

VPC IPAM added BGP route protection monitoring and delegated RPKI for BYOIP prefixes, enabling centralized visibility into RPKI validity, ROA strength, and route overlaps across accounts and Regions. After a one-time setup with an RIR, IPAM can automatically create and renew ROAs for BYOIP provisioning and on-premises prefixes, reducing manual effort and reliance on third-party tools. The feature is generally available in commercial Regions except AWS GovCloud (US) and the AWS China Regions.

New research on NAT state manipulation details how a host behind the same NAT as a victim can tamper with connection-tracking to hijack TCP sessions, spoof DNS responses, disclose externally mapped ports, or exhaust NAT connection tables. The NatJack research observed affected behavior across Windows and Linux, with CVE-2026-56181 (Windows Hyper-V NAT, CVSS 8.3) and CVE-2026-63913 (Linux Netfilter conntrack, CVSS 8.2) assigned for implementation-specific flaws. Vendor fixes are available; recommended mitigations include applying updates, segregating untrusted workloads from trusted systems sharing NAT, encrypting internal traffic, and using IP Source Guard where appropriate. No public evidence of in-the-wild exploitation was noted as of August 7, 2026.

Critical Fixes: From BI Platforms to Load Balancers and CMS

Metabase confirmed active exploitation of a maximum-severity flaw enabling unauthenticated remote SQL injection against its application database. The Metabase zero-day allows attackers to gain administrator-level access, modify configuration, steal stored credentials for connected databases, export data, and establish persistence. Metabase Cloud is patched; self-hosted users should upgrade immediately to fixed versions (x.58.24, x.59.21, x.60.17, x.61.11, x.62.9, x.63.5) or temporarily block “/api/session/reset_password,” then revoke sessions, review API keys and admin accounts, rotate database credentials, and examine logs. Indicators include a POST to “/api/session/reset_password” returning 400 followed by a GET to “/api/user/current” returning 200.

CISA added a critical command injection in Progress Kemp LoadMaster to the KEV catalog after active exploitation. The Kemp LoadMaster CVE (CVE-2026-8037, CVSS 9.6) arises from improper input handling in an escape_quotes() function and permits unauthenticated command execution. Telemetry recorded 792 exploitation attempts over 41 days from 65 IPs across 18 countries, with recent activity observed on August 4, 2026. Federal agencies are directed to patch by August 10, 2026, and all organizations should prioritize remediation and review for indicators of compromise.

WordPress fixed a high-severity pre-auth reflected XSS (CVE-2026-64638, CVSS 8.9) on the login page, backported through the 4.7 branch. The WordPress XSS can lead to attacker-controlled DOM elements that interact with user-profile.js, potentially steering same-origin REST requests; researchers showed a chain (XSS2Shell) to PHP code execution requiring administrator interaction, while WordPress emphasized the social engineering prerequisite. Immediate updates are recommended; there were no reported in-the-wild exploits as of August 7.

N‑able reported exploitation of a zero-day in its RMM server tied to incomplete remediation of a prior flaw. The N‑able N‑central issue (CVE-2026-18577, CVSS 8.2) enabled authentication bypass and account takeover in versions prior to 2026.3.1.7; attackers gained administrative access, abused the Take Control feature to access managed devices, and created persistence via a Cloudflare Tunnel service. On‑premises customers should update to 2026.3.1.10 and apply Hotfix 2, use the provided IoC lists and service template for checks, and conduct comprehensive log and account reviews; CISA has flagged both related vulnerabilities as actively exploited.

Tencent Zhuque Lab disclosed a long-standing use-after-free in Linux SCTP (CVE-2026-64564, “SCTPhantom”) introduced in Linux 2.6.25. The Linux SCTP flaw was turned into a local root exploit and, in lab tests, a container escape; fixes landed in 7.1.6, 6.18.42, 6.12.101, and 6.6.148 on August 3. Exposure is limited to local scenarios where SCTP is reachable, and some reproduction details remain unconfirmed; administrators should apply distribution updates or remove the SCTP module where not required and verify backported patches.

Breaches, Social Engineering, and Supply Chain Exposure

Healthcare financial software vendor Unlimited Technology Systems disclosed a breach affecting 3.8 million people. An unauthorized actor accessed files between October 5–10, 2025; the activity was discovered on October 19. Notices began July 1, 2026, with formal disclosure on July 20. Potentially exposed data include names, Social Security numbers, dates of birth, contact details, government IDs and insurance cards, intake forms, policy numbers, claims and benefits information, medical record numbers, dates of service, and diagnosis information. The company notified law enforcement and offered identity monitoring via Kroll.

In the UK charity sector, around 1,500 organizations using Beacon’s CRM were potentially affected by an incident disclosed on August 4, 2026. The Beacon incident involved a stolen access key and likely exfiltration of customer-held personal data (names, emails, telephone numbers, donation records); customers are advised to assume all stored data and attachments were downloaded. Beacon reports the platform does not store sensitive patient records, payment card details, or bank account information. Customers have been notified and advised on incident response steps, including potential reporting to the UK ICO.

Google Threat Intelligence Group and Mandiant detailed an extortion campaign run by UNC6671 that uses vishing to employees’ personal phones, directing victims to adversary-in-the-middle portals to intercept credentials and MFA tokens. The UNC6671 vishing activity enables session persistence by registering attacker-controlled MFA devices and facilitates lateral movement across SaaS platforms such as Microsoft 365 and Okta. Operating under brands including Redact, Pink, Helix, and Falcon, the group has been linked to over $10.6 million in tracked Bitcoin payments between January and May 2026, with initial demands exceeding $3 million and many settlements around $750,000. Recommended mitigations include phishing-resistant MFA, SSO with session controls, restricting authentication to managed devices and trusted networks, monitoring IdP logs for suspicious MFA registrations, and tooling to detect credential entry on unauthorized domains.

Supply chain risk also hit AI development environments. On March 24, 2026, TeamPCP poisoned the popular Python package LiteLLM on PyPI, using a .pth file to auto-execute code and harvest secrets such as cloud tokens and SSH keys. The Python package security report notes the malicious releases were public for about three hours, reaching tens of thousands of environments; researchers also observed a broader rise in malicious open-source packages and targeting of AI libraries. Recommended controls include pinning dependencies with checksum verification, auditing post-install hooks and auto-execution mechanisms before deployment to developer machines, and rotating credentials with provider log reviews after suspected exposure.

Research presented at Black Hat USA 2026 showed how allowed HTML and CSS in email can cross message boundaries and interact with webmail interfaces. The CSS webmail attacks demonstrated chains across multiple providers with outcomes including password capture, token exfiltration, UI hijacking, and manipulation of AI tools that process email. Some vendors issued partial fixes, while several vectors remained effective at publication. Suggested mitigations include sandboxing HTML emails in isolated iframes, restricting CSS and dangerous selectors, validating allow lists, checking for CSS “gadgets” before permitting custom attributes, blocking select menus and attacker-controlled image requests, and avoiding allow-listing risky domains.