Midjourney announced a new division, Midjourney Medical, and its first hardware product, the Midjourney Scanner — a full-body ultrasonic computed tomography device built around roughly 358,000 ultrasound transducers across 40 imaging modules. The device lowers a person into a shallow water-filled ring, firing soundwaves through the body from every angle to reconstruct cross-sectional images of muscle, fat, bone, and organs in about 60 seconds, with no radiation or magnets involved.
The underlying imaging hardware is licensed from Butterfly Network, a Nasdaq-listed ultrasound-on-chip maker whose shares jumped roughly 33% on the news. Midjourney's plan calls for a flagship "Midjourney Spa" opening in San Francisco's Union Square by 2027, with a longer-term goal of 50,000 scanners worldwide delivering a billion scans a month by 2031.
It is an unusual leap for a company whose reputation rests entirely on generative art, and the device has no FDA clearance for diagnostic use yet. Still, the announcement signals a broader pattern of frontier AI labs pushing capital and engineering talent out of software and into regulated, physical-world categories.
ChatGPT held over half the global AI assistant market as recently as January 2026. By the end of May, that figure had slipped to 46.4%, according to Sensor Tower — the first time it has dropped below the halfway mark since the category's creation in late 2022. Google's Gemini has climbed to 27.7%, while Anthropic's Claude has reached 10.3%, with the highest paid-subscription conversion rate of any major assistant.
The decline is relative, not absolute — ChatGPT remains the largest assistant by a wide margin, with over 1.1 billion monthly users, the fastest any app has reached that milestone. But the trend line has moved in one direction for eighteen straight months, even as OpenAI reportedly filed a confidential S-1 toward a public listing.
Sensor Tower's report frames this less as decline than as market maturation: total time spent across AI assistant apps is projected to double year-over-year, even as the field fragments across more competitors.
The US Commerce Department's Bureau of Industry and Security ordered Anthropic to suspend access to Fable 5 and its underlying foundation model, Mythos 5, for any foreign national — regardless of physical location, including Anthropic's own staff. Because the company has no reliable way to verify nationality at the application layer, it disabled both models globally within hours, citing national security concerns the government has so far described only verbally.
The stated trigger, per Anthropic's own account, involves a narrow technique for prompting the model to read a codebase and identify software vulnerabilities — a capability the directive treats in the same regulatory category as weapons systems. Anthropic disputes the characterization and is working to restore access; other Claude models remain unaffected.
The episode lands alongside a separate, ongoing dispute between Anthropic and the Pentagon over safety guardrails, leaving the company simultaneously contesting one federal restriction in court while complying with another.
In a nearly hour-long interview, Midha lays out what he calls "outputmaxxing" — the systems, habits, and organizational design choices that let frontier AI companies sustain maximum execution velocity without burning out their teams. The conversation moves from his early career through the rounds he's led in some of the most closely watched labs in the industry.
The discussion is notable less for any single headline than for its texture: a practitioner's view of how capital, talent, and research velocity actually interact inside the handful of labs racing toward frontier capability, at a moment when government policy, market share, and compute costs are all shifting simultaneously.
Salesforce confirmed a new AI agent partnership this week, deepening its push to embed autonomous agents directly into core CRM and sales workflows. The move arrives the same week federal AI procurement has split sharply along vendor lines, with one major lab's models pulled from government use while another's are fast-tracked into classified deployments.
Taken together, the pattern suggests enterprise AI buyers are increasingly weighing a vendor's regulatory alignment alongside raw model capability — a dynamic that didn't meaningfully exist in procurement conversations even a year ago.
Amazon S3 now supports annotations: structured, mutable context attached directly to objects, with up to 1,000 named annotations per object, each as large as 1 MB, for a combined ceiling of 1 GB per object in JSON, XML, YAML, or plain text. Unlike legacy object tags or upload-time metadata, annotations can be modified or deleted independently at any time without rewriting the underlying object.
When indexed into S3 Metadata's managed Apache Iceberg tables, annotations become queryable at scale via Amazon Athena or natural-language queries through an S3 Tables MCP server — letting AI agents and analytics tools discover relevant data across petabytes of objects without retrieval charges, even in S3 Glacier.
The launch reflects a broader infrastructure shift: as AI workloads strain conventional metadata systems, cloud providers are rebuilding storage primitives specifically around what autonomous agents need to find and act on data.
A new analysis of YouTube's "Get Ready With Me" category finds that LTK-affiliated creators lead share of voice in the format, with the bulk of that presence coming through unpaid, affiliate-driven video descriptions rather than purchased ad placements. The pattern suggests brand visibility in high-engagement formats increasingly runs through creator commerce networks rather than direct media buys.
The same data points to AI visualization products and community-driven playbooks — exemplified by Hinge's approach — leaning further into creator-native distribution rather than conventional paid media, as platforms reward authentic-feeling content over branded placements.
Washington's latest policy framing centers AI access and fairness as a core regulatory concern, even as consumer hardware makers race ahead with new form factors. Snap is testing AI-powered glasses aimed at bringing assistant-style interaction into a wearable, camera-equipped device, joining a wearables category already contested by Meta and several startups.
Separately, Amazon continues development of world models — AI systems trained to reason about physical space and object interaction rather than just text or images — alongside enhancements to its AI-powered email tools, underscoring how the largest cloud and retail platforms are pushing AI capability simultaneously into hardware, logistics, and productivity software.