"If you needed further evidence that AI comes first in pretty much everything nowadays, look no further than this year's Google Cloud Next show, which happened last week," systems editor Tobias Mann said on The Kettle podcast.
Tensor 8th‑generation split: separate chips for inference and training
Among the announcements at Google Cloud Next was a notable change to Google's Tensor chip lineup: Google introduced two versions of its 8th‑generation Tensor hardware, one optimized for inference and a different version optimized for training. The split signals a deliberate engineering choice to separate workloads that serve real‑time responses from those used to build and refine models.
The source reporting this, The Kettle podcast, presented the split as one of the clearer signs that AI now drives design decisions across the cloud stack rather than being a peripheral feature of product roadmaps.
Vertex AI becomes Gemini Enterprise Agent Platform
Google also retooled its developer surface: Vertex AI has been renamed and redesigned as the Gemini Enterprise Agent Platform. The platform was described as a single space intended to create and manage AI agents, consolidating tools and workflows for agent development and lifecycle management.
That renaming and redesign were presented at Google Cloud Next as an effort to provide a unified environment for enterprises building AI agents, reflecting how vendor product strategies are being recast around agentic AI capabilities.
New AI security agents shown off at Google Cloud Next
Security was part of the product narrative: Google Cloud Next included demonstrations of new AI security agents. The event positioned those agents alongside compute and development announcements, underscoring that security tooling is being integrated into the AI product story.
The Kettle conversation framed these security agents as part of the same trend tying hardware, platform, and protection into an AI‑first set of offerings.
Anthropic's Mythos: rapid improper access and early Project Glasswing results
Outside of Google Cloud Next, The Kettle discussed a separate, pressing incident: an "incredibly small amount of time" was required for someone to gain improper access to Anthropic's Mythos cybersecurity AI model. That breach prompted further investigation and testing.
Early results from Project Glasswing were also cited on the podcast; those results suggest Mythos is "less capable than the hype made it out to be." The combination of a swift unauthorized access event and Glasswing's early assessments framed Mythos as both a target and a subject of dampened expectations.
How technologists, procurement leaders, and end users may react
- Technologists and security teams: The dual announcements — new AI security agents at Google Cloud Next and the Mythos access incident — point to increased emphasis on building security into agent platforms and chip stacks. The discussion on The Kettle implies teams will need to reconcile rapid platform change (Gemini Enterprise Agent Platform) with the operational realities of preventing quick improper access events.
- Enterprises and procurement leaders: With Vertex AI renamed and consolidated into the Gemini Enterprise Agent Platform, procurement processes may shift toward platform‑level decisions for creating and managing agents rather than assembling disparate toolchains. The Mythos episode highlights the risk calculus around adopting third‑party AI models touted for cybersecurity functions.
- End users and the general public: The podcast framed both the product push at Google Cloud Next and the Mythos incident as part of a broader narrative: AI is now central to products, architectures, and security discussions. Public expectations about capabilities may be tempered by Project Glasswing's early findings that Mythos underperformed its hype.
The record from Google Cloud Next, as discussed on The Kettle, is simple and stark: AI is now the organizing principle for cloud hardware, platforms, and security tooling. At the same time, the Mythos episode serves as a reminder that aggressive platform repositioning and real‑world testing can yield very different stories — rapid access incidents and early independent assessments may undercut marketing claims. For practitioners and buyers, the twin messages from the event and the incident are clear: design choices are converging on AI agents, and security must move at the same pace.
Listen to The Kettle episode referenced here on Spotify or Apple Music, or follow the original writeup at The Register: https://go.theregister.com/feed/www.theregister.com/2026/04/27/google_cloud_next_proves_what/
