AWS Security and Identity Upgrades; Google, Cloudflare Ship

Major cloud providers concentrated on security, identity, and data platforms today, with AWS rolling out guardrails and connectivity features across services, Google making native graph analytics generally available in BigQuery, and Cloudflare unveiling a new bot-defense engine. Investigators also detailed ransomware and social engineering activity, and Microsoft worked to resolve a significant Exchange Online outage.

Identity, Authentication, and Safer Defaults on AWS

Amazon Redshift now supports authentication via AWS IAM Identity Center for both provisioned clusters and serverless workgroups when enhanced VPC routing (EVR) is enabled. Authentication and authorization traffic stays within the customer VPC over AWS PrivateLink, aligning with requirements for data residency, network isolation, and regulatory controls that prohibit public egress. The release also supports multi-Region replication for deployments where Redshift and the primary Identity Center instance are in different Regions.

Amazon Cognito introduced the GetClientToken API to support machine-to-machine authorization without configuring a user pool domain. App clients can authenticate with client ID and secret to obtain access tokens scoped to custom permissions, integrating with AWS SDKs and CLI and compatible with protections and network controls such as AWS WAF and VPC interface endpoints. The traditional OAuth 2.0 client-credentials flow via a domain remains supported.

AWS Lambda recursive loop detection is now available in all commercial Regions and enabled by default. The capability automatically identifies and stops unintended invocation loops between Lambda functions and supported sources like Amazon S3, Amazon SQS, and Amazon SNS. When a loop is detected, Lambda halts processing of the offending event and issues an AWS Health Dashboard notification with remediation guidance; customers who rely on intentional recursion can opt out per function via the PutFunctionRecursionConfig API.

Elastic Beanstalk added automatic Active Directory domain join for Windows Server platform versions released on or after August 18, 2026. Instances join the domain at boot, before application deployment, enabling Windows-integrated authentication, group policy application, and access to domain resources. Administrators can target organizational units and benefit from predictable computer names; deployments proceed even if a join fails, with the environment reporting the issue.

Automated Security Response on AWS gained an AI-driven Remediation Toolkit for generating custom actions with guardrails, plus automatic remediation for findings from Amazon Inspector, Amazon GuardDuty, and Amazon Macie. An updated console centralizes scoping and configuration by account, OU, Region, and tags, validates more than 100 controls, and reduces manual setup. Multi-channel notifications now include adapters for Email, Slack, Jira, and ServiceNow with severity filters, links, IaC snippets, and deadline enforcement.

Resilience and Cross-Cloud Connectivity

AWS Interconnect – multicloud entered public preview for private connectivity with Microsoft Azure. The product streamlines provisioning of resilient, scalable cloud-to-cloud links using an open interoperability specification, joining existing generally available integrations with Oracle Cloud Infrastructure and Google Cloud. The preview is available in selected Regions and can be provisioned via the console, CLI, or API.

Amazon Connect Global Resiliency now supports cross‑region routing across two active Regions, sending incoming contacts to the longest‑available matching agent in either location. Operations teams gain unified analytics and contact search across the pair, while administrators retain control over traffic distribution. Supported pairings include US East (N. Virginia) ↔ US West (Oregon), EU (Frankfurt) ↔ EU (London), and Asia Pacific (Osaka) ↔ Asia Pacific (Tokyo); enablement requires contacting an AWS representative.

Data Platforms and Agent Foundations

Amazon DocumentDB introduced in‑place major version upgrades directly from engine versions 3.6 and 4.0 to 8.0, preserving data and cluster configurations. The path simplifies migrations, reduces maintenance windows, and delivers the 8.0 feature set, performance enhancements, and updated security patches. AWS advises testing in non‑production and reviewing guidance on version differences and support timelines.

BigQuery Graph reached general availability, bringing native, ISO‑standard GQL to BigQuery alongside SQL without moving data. Google cites performance gains up to 2x for GQL and improvements for undirected traversals and cycle detection, plus new modularity via the CALL statement. The release integrates with BigQuery ML and Gemini, supports cross‑cloud virtual graphs via Unity Catalog, AWS Glue, or Snowflake, and targets use cases such as threat and fraud detection, digital twins, and AI grounding.

AWS Agent Registry is now generally available as a private, governed catalog for agents, tools, MCP servers, and custom resources. GA adds enterprise capabilities including infrastructure‑as‑code (CloudFormation, Terraform, AWS CDK), tagging, cross‑account and cross‑organization sharing via AWS RAM, and auto‑detection of AgentCore runtimes and gateways. The service integrates with Amazon Bedrock AgentCore, Amazon Quick, and Kiro IDE, and exposes an MCP server interface for IDE queries; preview audit features such as approval workflows and CloudTrail trails persist.

Amazon Quick now integrates with AWS Agent Registry, allowing business users to discover and enable MCP servers and agents directly in the Quick UI with connection details auto‑populated from the registry. Enabled assets can be shared across teams for chat, agents, apps, flows, and deep research, improving access governance and reducing duplication. Administrators can connect their registry from the Quick admin console.

Threat Activity and Service Disruptions

Berlin’s municipal government confirmed data theft following a Rhysida ransomware intrusion, with exfiltration claims of about 5.79 TB encompassing administrative, financial, HR, health, and mapping records, among others. Investigations led by state and federal authorities continue, and officials report no evidence that election data or systems for the upcoming vote were affected; the city will not pay the ransom. Details and timeline are provided by BleepingComputer.

Analysts from CloudSEK and Gambit Security reported that Aurora ransomware operators used AI coding assistants, chiefly Cursor and an agent with Anthropic’s Claude Sonnet, to aid intrusions against more than 20 organizations between April and July 2026. The group’s tooling spans Windows and Linux/ESXi encryptors from a Zig codebase and scripts for discovery, lateral movement, and abuse of AD CS, with initial access including email bombing and phone‑based social engineering. Financial analysis indicates affiliate payout splits ranging from 54% to 79%. See The Hacker News for details.

Unit 42 documented “Spring Ring,” a coordinated campaign using external Microsoft Teams accounts and live voice calls to impersonate corporate IT support and deliver remote access tooling. Two related operations attempted persistent footholds, extension sideloading, lateral movement, and, in one case, a PetitPotam NTLM relay against domain controllers; detections blocked payload execution and domain‑takeover attempts. The analysis underscores risks from trusted collaboration platforms and vishing. Read more at Unit 42.

Microsoft investigated a significant Exchange Online outage (incident EX1464935) that began at 5:30 PM UTC and caused email delivery failures, authentication errors, and administrative disruptions for many customers. Engineers isolated a common failure pattern tied to authentication and protocol connectivity and worked on remediation while monitoring service telemetry. Coverage is available via BleepingComputer.

Cloudflare introduced Adaptive Intelligence, a bot‑detection engine that continuously retrains on live traffic, deploys disposable rules to disrupt attacker learning, and mines labeled signals across millions of sites to target low‑rate, distributed campaigns. Candidate detections are validated against recent traffic, rolled out gradually as inputs to the bot score, and can be paused or rolled back to avoid impacting legitimate users; the approach complements existing behavioral validation.