TLDR IT 2026-07-29

77% of organizations say AI adoption is outpacing their governance capabilities (IBM, 2026 Tech Leader Study). Employees are shipping code faster than IT can track it. Tines 3B gives every team the power to build with AI, and gives IT the control to run and monitor that work safely. With protected credentials, monitoring, and full auditability, Tines 3B lets IT teams shift from blockers to enablers. Get started today >

On August 19th, hear from Headspace on how they're governing AI at scale with 3B. Register now.

Snowflake's Cortex AI Gateway gives organizations centralized control over how agents access models. The product is designed to improve governance and prevent uncontrolled usage from creating unexpectedly high AI costs.

Microsoft's Project Perception is an AI cybersecurity system that coordinates red, blue, and green team agents to defend against AI-driven attacks. Entering public preview August 3, it is based on Microsoft's new MAI-Cyber-1-Flash model used with OpenAI's GPT-5.4 for the hardest tasks. Microsoft says the combination scores 96% on CyberGym and is initially available only to MDASH customers.

The Model Context Protocol's July 28 revision introduces major architectural changes aimed at supporting large-scale enterprise agent deployments. The update moves MCP beyond its earlier stateful design, but could require existing implementations to adapt.

SAP argues that enterprise agents need governed knowledge graphs rather than simple access to disconnected documents and applications. Reliable agents require shared business context, relationships, and permissions that can be consistently applied across workflows.

Zero Trust is insufficient for AI agents because verifying identity does not stop attackers from hijacking behavior through malicious language in context. Zero Custody holds that the reasoning loop should hold nothing that matters, so a compromised agent has empty hands. Secrets and raw values should stay with external brokers or gateways that inject or re-identify them only after policy checks.

A survey of 101 enterprises found that companies are building infrastructure to supply agents with business context faster than they can ensure that context is trustworthy. The next enterprise AI bottleneck may be data lineage, permissions, and accuracy rather than retrieval performance.

Microsoft's Project Perception combines specialized and frontier models with shared security context and agents that can act across its security products. Microsoft says the architecture can improve reasoning consistency while cutting costs by nearly 50%. The public preview begins on August 3.

Dell PowerStore can now serve as external storage for Nutanix Cloud Infrastructure 7.6, with deployment and lifecycle management integrated through Dell Private Cloud and Nutanix Prism. The integration gives customers another path away from legacy virtualization stacks while allowing compute and storage to scale independently.

Cloudflare is open sourcing pvcli (privacy-client), a CLI for debugging privacy-preserving protocols such as Oblivious HTTP, under the Apache-2.0 License. The tool automates binary HTTP encoding, encryption, and multi-party relay/gateway flows that previously required manual bit-level work and one-off custom clients. It logs each protocol step so developers can isolate failures across relays, gateways, and origins.

Kioxia's new NX1 SSD is designed for GPU servers and hyperscale data centers where storage performance and heat management are becoming more important. The E1.S drive supports capacities up to 15.36 TB and offers improved write performance over the company's previous generation.

Security researchers uncovered two AI-built Phishing-as-a-Service kits, TokenLover and YaksaLover, that automate business email compromise against Microsoft 365. Marketed like legitimate SaaS products with operator dashboards, they industrialize documented techniques such as device code phishing and Windows Hello persistence. One kit uses AI to map payment flows from harvested mailboxes, and both can keep access after password resets.

AI-generated code often omits production essentials such as error handling, rate limiting, and instrumentation, and observability skills let agents verify work against runtime behavior to feed observations back into the next plan.

Meta argues that faster metadata pipelines, ingestion systems, and GPU-adjacent flash can reduce expensive idle time in AI clusters.