Cerebras stock nearly doubles on day one as AI chipmaker hits $100 billion

Cerebras Systems saw its share price almost double on its first day of public trading, valuing the company at $100 billion — a milestone that signals sustained investor confidence in dedicated AI chip infrastructure despite pressure from efficiency-first models. The company’s wafer-scale architecture targets inference-heavy workloads where GPU bandwidth becomes the constraint.

Enterprise AI agent authorisation is broken — and successful authentication makes it worse

Cisco researchers presenting at RSAC 2026 found that enterprises can confirm an AI agent's identity, but cannot reliably control what it is permitted to do once authenticated. The gap between authentication and authorisation is wide enough for agents to act on permissions they were never meant to exercise.

Developers can now debug and evaluate AI agents locally with Raindrop's open-source Workshop

Raindrop has released Workshop, an open-source local debugger and evaluation suite for AI agents. Teams can inspect agent calls, trace reasoning chains, replay sessions, and run standardised evaluations without sending data to a remote service — closing the observability gap that has blocked enterprise governance teams from trusting agent outputs.

Claude Code's /goals command separates the agent that decides it's done from the one that does the work

Android will know what you want before you do — and enterprise IT should plan for it

Google is rolling out contextual AI suggestions on Android that learn from user patterns — calendar activity, messaging habits, app usage — to surface actions before users request them. For organisations managing BYOD or corporate-owned Android devices, the feature raises questions about what employee work patterns the AI layer is observing and how that data is scoped.

Qualcomm's Snapdragon X chips are expanding beyond Windows with the Googlebook platform

Qualcomm's Snapdragon X chip family, previously exclusive to Windows Arm laptops, is now confirmed for Google's Googlebook platform. The expansion creates a second major consumer Arm platform targeting the same enterprise hardware refresh cycle that Apple silicon has dominated for three years.

Google is quietly turning Gemini into a productivity OS — and enterprise buyers should take note

Google's Gemini is being repositioned from standalone AI assistant to an operating layer across Google Workspace, Android, and Chrome. New integrations let Gemini act on email, calendar, documents, and device functions with minimal friction — narrowing the gap with Microsoft Copilot as an enterprise AI productivity platform.

Android will automatically end calls that match known bank scam patterns to protect users

Google is rolling out an Android feature that automatically hangs up calls matching known bank-scam patterns, using on-device AI to detect social engineering attempts in real time without sending audio to Google servers. The feature targets phone-based fraud that bypasses perimeter security controls.

PCIe 8.0 is fast, but it may require enterprise data centres to replace server connectors

The PCIe 8.0 specification doubles bandwidth over PCIe 5.0, but signal integrity requirements at those speeds may necessitate a new physical connector standard. Enterprise data centre managers planning multi-year hardware refreshes need to assess whether existing cabling and backplane infrastructure will be compatible with the next generation.

Stop depending on Claude and ChatGPT — use them to build tools your organisation actually owns

As AI assistants embed deeper into enterprise workflows, practitioners are raising a structural concern: building critical processes around proprietary AI platforms creates dependency on tools whose features, pricing, and behaviour can change without warning. The more defensible position, the argument goes, is to use AI to build durable, self-owned tooling rather than treating the AI platform itself as the product.