🔥 The Week in AI News - Jun 1 - 5, 2026

Apple, OpenAI, Meta, Higgsfield, ImagineArt, ElevenLabs, Black Forest Labs, Google, Codex, ChatGPT, Meta, WhatsApp, Microsoft, TSMC, Anthropic, Anthropic, Google, Meta, Google Photos and more.

▪️ Mira Murati stepped back into the spotlight with a careful vision for Thinking Machines Lab.

Mira Murati made one of her first major public appearances since leaving OpenAI, and she used it to preview a more fluid kind of AI interface. The bigger signal was not just the product direction, but the way another major AI founder is trying to define the next layer after chat.

▪️ Apple approved Poke as the first AI agent on Messages for Business.

Poke became the first standalone AI agent approved for Apple’s Messages for Business platform, bringing agent-style help into iMessage. That matters because the next wave of AI may not feel like opening an app — it may feel like texting something that can actually do the task.

▪️ OpenAI pushed Codex further into white-collar work.

OpenAI launched new Codex tools aimed at knowledge workers, with plug-ins for jobs like data analysis, sales, creative production, product design, investing, and banking. The story here was simple: coding agents were no longer just for developers, they were becoming workplace agents.

▪️ Meta rolled out an AI creator assistant on Facebook.

Meta introduced a new AI assistant to help Facebook creators understand performance, brainstorm content, and ask follow-up questions about their audience. It was another sign that creator platforms are turning analytics dashboards into conversational strategy tools.

▪️ Higgsfield brought its AI tools into Figma.

Higgsfield’s Figma plugin went live, putting AI creative tools closer to where designers already work. That kind of integration matters because the winning AI tools are not just flashy standalone apps anymore — they are the ones that disappear into the workflow.

▪️ Higgsfield moved directly inside Premiere Pro.

Higgsfield added Reframe, Remove BG, Upscale, Draw to edit, and Edit Video inside Premiere Pro, cutting out the export-and-switch routine. For video creators, this was the kind of update that turns AI from a side tool into part of the editing timeline.

▪️ ImagineArt launched Ads Studio for commercial creative production.

ImagineArt introduced Ads Studio as part of a broader push into brand and marketing workflows. The move fit the bigger pattern of AI image platforms trying to become full creative production systems instead of single-output generators.

▪️ ImagineArt added Simple Mode to Workflow.

ImagineArt introduced Simple Mode for Workflow, letting users describe what they want instead of building everything through nodes. That was a smart move because powerful workflow tools only become useful at scale when nontechnical creators can actually move through them.

▪️ ElevenLabs launched Flows Agent inside ElevenCreative.

ElevenLabs added Flows Agent as a conversational co-editor for ElevenCreative Flows. Instead of manually wiring every node, creators could describe the campaign or asset flow they wanted and let the agent start building the structure.

▪️ ElevenLabs brought Hasbro character voices into its Iconic Marketplace.

ElevenLabs added rights-cleared Hasbro voices, including characters like Cobra Commander, Megatron, Col. Mustard, and Optimus Prime. This was a big IP moment because licensed character voices are a very different lane from generic AI voice generation.

▪️ Martin Scorsese became an unlikely Hollywood voice for AI.

Martin Scorsese signed on as a partner and adviser to Black Forest Labs, with the technology positioned around storyboarding rather than full production replacement. The move mattered because it showed how even AI-skeptical creative industries may start accepting the technology when it supports the creative process instead of trying to replace it.

▪️ Google tested a floating desktop AI search bar.

Google tested a floating AI Search bar in Chrome Canary that opened as a standalone window on the desktop. It pointed to a future where AI search may become more like a system-level command center than a website you visit.

▪️ WWDC 2026 built anticipation around Siri and Apple Intelligence.

Ahead of WWDC, expectations centered on a more conversational Siri, deeper Apple Intelligence features, and possible support for AI agents. The pressure on Apple was obvious: the company needed to show that its AI strategy could finally move from careful promise to useful product.

▪️ ChatGPT’s upgraded memory rolled out more broadly.

OpenAI expanded ChatGPT’s upgraded memory system, giving the assistant a better way to remember preferences and update them over time. This mattered because memory is one of the features that turns AI from a tool you prompt into a system that starts feeling more personal.

▪️ Meta made its WhatsApp Business AI agent available globally.

Meta expanded its Business Agent for WhatsApp globally, giving businesses AI help for customer questions, product recommendations, appointments, and lead qualification. WhatsApp has always been a messaging layer for business, and Meta was clearly trying to make it a workflow layer too.

▪️ Microsoft launched Scout as an OpenClaw-inspired personal assistant.

Microsoft introduced Scout, an always-on agentic assistant built for the Microsoft 365 ecosystem. The bigger idea was persistent work memory: an AI assistant that learns your patterns, builds skills over time, and becomes harder to leave once it understands how you operate.

▪️ Microsoft introduced a better way to control AI agent behavior.

Microsoft released the Agent Control Specification, an open source standard for defining what agents can and cannot do. As AI agents get more access to tools and systems, the boring layer of control, approval, and audit trails is becoming one of the most important parts of the stack.

▪️ Microsoft launched ASSERT for AI behavior testing.

Microsoft introduced ASSERT, a tool that turns plain-language descriptions of expected AI behavior into structured tests. That was important because companies do not just need smarter agents — they need repeatable ways to prove those agents are behaving correctly.

▪️ ChatGPT reportedly reached 1 billion monthly active users faster than any other app.

ChatGPT reportedly crossed 1 billion monthly active users roughly three years after launch. That kind of adoption made it clear that AI was no longer an emerging software category — it was becoming a default digital behavior.

▪️ Meta scaled back an employee tracking tool used for AI training.

Meta reportedly adjusted its Model Capability Initiative after internal backlash over staff activity tracking. The update showed the tension inside AI companies themselves: everyone wants better training data, but not every data source feels acceptable when humans are the ones being watched.

▪️ TSMC struggled to keep up with AI demand.

TSMC said customer demand remained so high that even its U.S. factory buildout would take a long time to fully support it. The AI boom is still software on the surface, but underneath it, the whole thing keeps coming back to chips, fabs, memory, power, and physical capacity.

▪️ Anthropic filed confidentially to go public.

Anthropic filed confidentially for an IPO after a massive funding round and a valuation reportedly near the trillion-dollar mark. The filing made the AI race feel even more like a public-market event, not just a private startup battle.

▪️ Anthropic scaled Claude Mythos into critical infrastructure security.

Anthropic expanded Project Glasswing, giving more organizations access to Claude Mythos for finding and fixing serious software vulnerabilities. This was one of the more consequential AI stories of the week because it moved frontier models into the security layer of power, healthcare, communications, and other critical systems.

▪️ Google rolled out fake call detection for AI deepfake scams.

Google launched fake call detection for Android to help identify scams that use spoofed numbers and AI-generated voices. As AI impersonation gets more convincing, phone security has to move from spam filtering to identity verification.

▪️ Microsoft made AI behavior testing easier with text descriptions.

Microsoft introduced another tool that lets developers spin up AI behavior tests using plain-language descriptions. That mattered because agent reliability is becoming less about building impressive demos and more about proving what the system will do when the stakes are real.

▪️ Publishers gained a new way to opt out of AI Search.

A new regulation gave publishers a clearer way to opt out of appearing in AI search experiences. This was a major media and platform moment because publishers have been asking for more control over how their content shows up inside AI-generated answers.

▪️ Meta built temporary data centers in tents to speed up AI infrastructure.

Meta reportedly built rapid-deployment data center structures in Ohio to speed up AI capacity. It sounded extreme, but it also captured the moment perfectly: the AI race is now moving so fast that even infrastructure is being forced into startup mode.

▪️ Google Photos prepared an AI-generated digital wardrobe.

Google Photos started rolling out an AI wardrobe feature that could use a person’s photo library to generate outfit combinations and virtual try-ons. It was another reminder that AI personalization is moving into very ordinary parts of life, including clothes, photos, and daily style decisions.

▪️ OpenAI reportedly moved toward combining Codex and ChatGPT.

OpenAI reportedly made plans to bring Codex and ChatGPT closer together. That direction made sense because the future of AI work probably will not be split between “chatting” and “doing” — the same interface will need to understand, build, revise, and act.