Claude Security Opens to All Enterprise Customers
If your organisation runs enterprise software and relies on Claude for development work, Claude Security is worth paying close attention to today. Anthropic has moved the tool out of closed preview and into a public beta available to all Claude Enterprise customers, giving security teams an AI-powered system that can scan entire codebases for vulnerabilities and generate targeted patches.
What sets this apart from generic SAST tools is the reasoning layer: Claude Security traces data flows, reads source code, and models interactions between components much as a skilled security researcher would, rather than matching patterns against a fixed ruleset. A partner ecosystem is already in place from day one, with CrowdStrike, Microsoft Security, Palo Alto Networks, SentinelOne, TrendAI, and Wiz embedding Opus 4.7 on the technology side, and services firms including Accenture, BCG, Deloitte, Infosys, and PwC helping enterprises deploy it. Support for Team and Max plan users is coming shortly.
GPT-5.5-Cyber Rolls Out to Critical Cyber Defenders via Trusted Access Programme
OpenAI has begun a controlled rollout of GPT-5.5-Cyber, a security-focused variant of its latest frontier model, through a new Trusted Access for Cyber programme. The model is not broadly available: access is being granted first to vetted government and critical infrastructure defenders, with CEO Sam Altman announcing the initial rollout directly via X ahead of formal press coverage.
For organisations working in cyber defence, the significance is that this is frontier-grade reasoning capability applied specifically to threat analysis and vulnerability research, rather than a general model being loosely repurposed. OpenAI has paired this with a five-pillar cybersecurity action plan covering democratised access to AI-powered defence tools, cross-government and industry coordination, and maintained oversight of how these capabilities are deployed. The timing is notable: Anthropic launched Claude Security in the same news cycle, and both companies are now competing directly on dedicated AI security products as a distinct category.
OpenAI Launches Advanced Account Security with Yubico Hardware Key Support
Alongside the GPT-5.5-Cyber news, OpenAI has launched Advanced Account Security: an opt-in set of protections for ChatGPT accounts that includes hardware security key support via a new partnership with Yubico. The feature is positioned at high-value individuals but is available to any user who wants it.
For developers and businesses managing API credentials or sensitive workflows in ChatGPT, hardware-backed authentication represents a meaningful security upgrade, bringing ChatGPT account security closer to the standards expected in enterprise software. Given the volume of sensitive work now flowing through ChatGPT in professional contexts, this feels overdue rather than premature.
Gemini Enterprise Agent Platform Takes Centre Stage at Google Cloud Next
Google's biggest enterprise AI push at Cloud Next '26 centred on a rebuilt platform now called the Gemini Enterprise Agent Platform. If you have been watching Vertex AI and wondering when Google would make it easier to build, deploy, and govern agents at scale, this is the answer. The platform brings together model selection, agent building, orchestration, security, and DevOps into a single offering, with a new Agent Designer, an Inbox for managing agent activity, support for long-running agents, and Skills and Projects functionality designed for production environments.
Access to Gemini 3.1 Pro is included for complex workflow orchestration, alongside Gemini 3.1 Flash Image for visual asset generation and Lyria 3 for audio. On the infrastructure side, 8th-generation TPUs were confirmed, delivering nearly three times the compute performance of the previous generation. Google also committed $750 million to develop its partner network for agentic AI, with early partners including Salesforce and SAP. More than 260 announcements were made across the event.
DeepMind published its AI co-clinician research initiative on 30 April, outlining a vision for what it calls triadic care: AI agents supporting patients in their health journeys under the clinical authority of their physician. This is a research direction rather than a product launch, but it signals where DeepMind's healthcare AI work is heading. Google DeepMind Blog, 30 Apr
Microsoft 365 E7 and Agent 365 Are Now Generally Available
Today is the day Microsoft launches its most significant enterprise licensing tier in a decade. Microsoft 365 E7, priced at $99 per user per month, is now generally available and bundles together Microsoft 365 E5, Microsoft Entra Suite, Microsoft 365 Copilot, and Agent 365. The last of those is the most consequential piece for anyone building or governing AI workflows in a large organisation.
Agent 365 is a central control plane for observing, governing, managing, and securing every AI agent operating across the Microsoft ecosystem. It is available standalone at $15 per user per month or included within the E7 bundle. For IT and security leaders who have been trying to track sprawling AI agent deployments across their organisations without a unified view, this is the tool Microsoft has been signalling for months. The launch also includes pricing changes for Dragon Copilot across all geographies where it is available.
Industry Themes
The most striking pattern across today's announcements is the convergence of frontier AI on cybersecurity. In the same 24-hour period, Anthropic launched Claude Security for enterprise codebase scanning and OpenAI announced the rollout of GPT-5.5-Cyber to critical defenders. Both companies appear to be positioning AI not merely as a productivity layer but as active defensive infrastructure. For anyone building security workflows or evaluating enterprise AI tools, this is the clearest signal yet that dedicated security-focused AI products, rather than repurposed general assistants, are becoming a distinct product category with their own competitive dynamics.
The second theme is enterprise AI governance coming of age. Microsoft's Agent 365 and Google's Gemini Enterprise Agent Platform both made major moves in the same news cycle. The common thread is that enterprises are moving beyond asking whether to deploy AI agents and are now demanding centralised control planes, observability tooling, and governance frameworks for agents already in production. The era of experimental AI deployments is giving way to something that needs to be managed, audited, and secured at scale, and the vendors are responding with the infrastructure to match.
A third, quieter observation: the OpenAI GPT-5.5-Cyber rollout surfaced first via Sam Altman's post on X, ahead of formal press coverage. This is a pattern worth noting. OpenAI in particular increasingly uses X as a first channel for significant product announcements, which means monitoring official X accounts remains a meaningful advantage for staying ahead of the news cycle. Both [Direct] items in today's digest would have been missed or delayed without the social channel scan.