Google Cloud unveils AI updates on standards & security

Google Cloud has outlined a new set of artificial intelligence updates spanning data standards, security partnerships and model availability. The latest additions focus on enterprise AI deployment and governance.

In its latest monthly round-up, Google Cloud introduced the Open Knowledge Format, a specification intended to formalise the so-called LLM-wiki pattern into a portable standard for metadata, context and curated knowledge used by AI systems.

The format is vendor-neutral and designed for both software agents and human operators. The move reflects a wider industry effort to reduce fragmentation as companies connect large language models to internal data and external tools.

Another headline development was a collaboration with Apple tied to Apple's expanded Private Cloud Compute systems. According to Google Cloud, the work draws on its Confidential Computing portfolio and Titanium security architecture, placing privacy and protected processing at the centre of the arrangement.

The update also included the general availability of Claude Fable 5, Anthropic's latest model, on Google Cloud. That adds to a growing list of third-party models available through Google's AI stack as cloud providers compete to offer customers a broader choice of systems rather than relying only on their own models.

Retail was also part of the latest push. Google Cloud used the period to present Cloud Atelier, a virtual shopping experience built around product discovery, as it seeks to show how AI tools can be applied beyond core developer and infrastructure use cases.

Security focus

Much of the broader update was framed around AI security, reflecting both customer demand and the growing prominence of cyber defence in cloud competition. Google Cloud highlighted internal work by its security teams to shift towards what it described as a more autonomous and proactive model.

That work includes embedding AI agents into the software development lifecycle to create automated safeguards around code review and software protection. Google Cloud also pointed to guidance from Mandiant on red teaming AI applications and new commentary from Chief Information Security Officer Chris Betz on lessons learned while developing AI Threat Defence.

The emphasis continues a theme running through Google Cloud's announcements in recent months. Earlier updates included the launch of Google AI Threat Defence, described as an always-on autonomous security platform, along with the addition of security agents in Google Security Operations and expanded support for model and identity controls.

In previous months, Google also completed the acquisition of Wiz, the cloud and AI security platform whose brand it said it would retain. The deal gave Google Cloud a stronger position in multicloud security, an area that has grown in importance as companies spread workloads across different providers while trying to maintain a single view of risk.

Platform changes

The recent announcements build on a broader restructuring of Google Cloud's AI product line around its Agent Platform. Earlier this year, the company said future Vertex AI services and roadmap changes would be delivered through Gemini Enterprise Agent Platform rather than as a standalone service.

That shift signalled Google Cloud's intention to make AI agents the organising principle for a large part of its software portfolio. It also introduced a series of related tools, including Agent Designer, long-running agents, an inbox for monitoring agent activity, project-based memory controls, task shortcuts known as Skills, and an Agent Gallery for third-party integrations.

Alongside software changes, Google Cloud has used the year to add more infrastructure and model options. Previous updates included Gemini 3.5 Flash, Gemini Omni, Nano Banana 2 and Nano Banana Pro, Gemini Embedding 2, and Veo 3.1 Lite, as well as new TPU systems for training and inference.

Those additions underline the breadth of the company's AI strategy. Google Cloud is competing at several layers at once, from custom chips and model hosting to application-building tools, security controls and industry-specific services for areas such as retail and customer experience.

Open standards

The Open Knowledge Format is likely to attract attention because it addresses a persistent problem in enterprise AI: how to package internal knowledge in a way that different models and tools can use consistently. Many businesses have experimented with retrieval systems, knowledge graphs and document-based grounding methods, but interoperability has remained limited.

Positioning the format as open and portable may be intended to encourage adoption beyond Google Cloud's own products. That would echo a pattern seen elsewhere in AI infrastructure, where vendors often back open specifications to reduce customer concerns about lock-in while still hoping their own platforms benefit from broader usage.

The same logic appears in earlier efforts such as the Universal Commerce Protocol, announced at the start of the year for agentic retail, and in the company's support for managed model context protocol servers for developers building production AI agents.

Business pressure

For customers, the stream of releases shows how quickly cloud AI offerings are changing. Beyond headline model launches, Google Cloud has published a growing body of operational guidance on topics such as measuring the return on investment of AI-assisted software development, securing inference on Kubernetes, production-ready agent design and capacity planning.

That emphasis suggests the market is moving from experimentation towards deployment questions around cost, governance, reliability and security. While varied in subject matter, the latest update repeatedly returns to those themes as businesses weigh how far and how quickly to push AI into live systems.

One of the more notable signals is that Google Cloud is no longer presenting AI primarily as a set of discrete models. Instead, it is framing the market around standards, managed agents, security controls and integration with existing enterprise data and workflows.

The strategy aims to make Google Cloud relevant whether a customer wants to host a third-party model, build internal agents, secure AI-driven applications or structure company knowledge for machine use. In that sense, the latest announcements show a company trying to turn AI from a collection of tools into a full operating environment for enterprise computing.