Anthropic is expanding Claude Code from a single coding assistant into a coordinated system of cloud-based agents—and is now making that model more practical for regulated organizations. Its new Projects workflow lets a coordinator divide goals among parallel threads, while Coder’s Agent Relay runs Claude Code agents inside customer-controlled, sandboxed environments with network policies and audit logs.

Source

AWS is rolling out a new sign-up flow that gives new customers a ready-to-use project, automatic permissions configured by AWS or coding agents, and a monthly spending ceiling starting at $20. Reaching that ceiling pauses the project instead of allowing more charges, while AWS’s service terms warn that an unreopened project may eventually be closed and its content deleted.

Source

Satya Nadella joins All-In hosts David Friedberg, Jason Calacanis, and Chamath Palihapitiya to discuss AI safety, Microsoft’s model strategy, Azure infrastructure, and the economic future of artificial intelligence. Nadella argues that the industry should shift from chasing raw model power toward reliable, auditable systems that enterprises can control and replace.

Source

AWS says it cannot restore some customer data affected by the March drone strikes on its Middle East infrastructure. The loss covers the entire Bahrain region and one Availability Zone in the UAE, leaving customers dependent on backups, surviving copies, or migrations to other AWS regions.

Source

Oracle is cutting staff again even as its remaining performance obligations reach a record $664 billion, underscoring the financial pressure behind its aggressive AI data-center expansion. The latest layoffs follow a year in which Oracle reduced its workforce by about 13%, while continuing to borrow and raise capital to fund cloud infrastructure.

Source

Salesforce is expanding Agentforce across AWS and Google Cloud while introducing Enterprise AI Harness, a control layer for governing agents, models and enterprise data across multiple platforms. The integrations let agents use Salesforce context and workflows alongside AWS and Google Cloud services without forcing customers to move all their data or standardize on one model provider.

Source

Splunk is expanding in two directions at once: bringing AI-driven operations closer to sensitive on-premises data with Cisco and Nvidia, while moving beyond AWS integrations toward jointly designed, agent-based security products. The strategy positions Splunk as a full-stack observability and security platform spanning cloud, data center, and edge environments.

Source

A mass-scanning campaign is exploiting CVE-2026-39364 in internet-exposed Vite development servers to harvest AWS credentials, Azure tokens, environment files, and infrastructure configuration. F5 honeypots recorded more than 800 attacks and about 32,000 related events in one month, making publicly reachable development environments an immediate credential-rotation concern.

Source

Azure AI Speech LLM 2607 is Microsoft’s latest large language model update for Azure Speech-to-text, designed to make voice transcription more accurate, responsive, and easier to customize. It improves multilingual and mixed-language transcription without requiring application changes. The service-side update also adds a cleaner phrase-list mechanism for domain vocabulary and is available through the Fast API, Real-Time API, and Foundry Playground.

Source

Microsoft has begun the final phase of retiring Azure SQL Data Sync: subscriptions that have never used the service can no longer create new sync groups, while existing deployments have until September 30, 2027, to migrate. The deadline is firm, so administrators with bidirectional synchronization between Azure SQL, SQL Server, and on-premises databases should start assessing replacement architectures now.

Source

Microsoft is urging Azure architects to stop applying a blanket three-zone design to every production workload. Its new framework recommends deciding zone placement component by component, using Azure-managed redundancy where available, and reserving three zones for systems that genuinely need the extra failure domain, quorum, durability, or post-failure capacity.

Source

A financially motivated attacker compromised cloud infrastructure and used an AI coding chatbot plus agent instructions to build a credential-stealing operation in less than six hours. Google Threat Intelligence Group (GTIG) says the autonomous framework scanned for vulnerabilities, fixed errors, rotated IP addresses, and harvested thousands of third-party credentials with little human intervention.

Source

Perplexity CEO Aravind Srinivas says users increasingly want AI agents to run on their own hardware to strengthen privacy, cybersecurity, intellectual-property protection, and digital sovereignty while controlling costs. In an interview with CNBC’s Kate Rooney, he described Perplexity’s fully local DGX Spark deployment and a hybrid model that keeps sensitive data on-premises while still accessing cloud-based frontier models.

Source

Microsoft is shifting its AI agent cost strategy from measuring usage to controlling what each agent sees, retrieves, remembers, and can do. Building on its earlier Foundry cost-cutting playbook, the company now presents context engineering as a way to reduce token spending while improving agent accuracy over time.

Source

Google Cloud’s us-central1-b outage stemmed from a maintenance mistake that disconnected every fiber path serving the affected zone. The September 1 incident left virtual machines unreachable and caused severe network degradation for more than four hours, despite the region’s multilayer redundancy.

Source

Google is pushing a more flexible pricing model for AI agents as companies struggle to control rapidly rising usage costs. The plan combines pay-as-you-go billing, token discounts of up to 20%, hard monthly spending caps, and a zero-cost base subscription—directly challenging Anthropic’s recurring seat fees and Microsoft’s separate usage licenses.

Source

Amazon Quick now connects directly to Microsoft Outlook, allowing users to summarize threads, draft contextual replies, manage messages, and schedule meetings without leaving the mailbox. The extension can also trigger broader workflows, but its ability to take actions makes permissions, data handling, and response verification important considerations for IT teams.

Source

AWS has published an enterprise deployment pattern that routes OpenAI ChatGPT Codex through LiteLLM on Amazon ECS and Amazon Bedrock, giving teams centralized model access, budgets, rate limits, and usage visibility. The important detail is that Codex still runs its task loop and approved tools on the developer’s workstation; only model requests pass through the AWS-hosted gateway.

Source

AWS has expanded Amazon Elastic VMware Service (EVS) to 22 regions and doubled the supported environment size to 32 hosts, giving organizations a broader path for moving VMware workloads out of aging data centers. The service now supports VMware Cloud Foundation (VCF) 9.0 and 9.1 on Amazon EC2 bare-metal infrastructure, including a lift-and-shift VCF-to-VCF migration model that preserves existing VMware operations.

Source