Midjourney’s Body Scanner Shocks Health Tech, OpenAI Limits Science ChatGPT Access, Microsoft’s Copilot Hacked for Data Theft, Apple Hides AI Extensions in iOS 27

**SPEAKER_2** (0:05)
Welcome to Digimasters Shorts.

**SPEAKER_3** (0:07)
We are your hosts, Adam Nagus and Carly Wilson, delivering the latest scoop from the digital realm.
Midjourney CEO David Holtz has unveiled the company's first hardware product, an ultrasound-based full-body scanner, marking a sharp turn from the AI-generated images that made the company famous.
Dubbed the Midjourney Scanner, the device uses a ring of thousands of transducers to capture vertical slices of the body and generate detailed 3D images of muscle, fat, bone and organs. The system, developed with Butterfly Network, relies on 40 ultrasound-on-chip modules and two petaflops of processing power, with each scan taking about 60 seconds. Users step onto a platform that lowers them into water, where ultrasonic waves pass through the body from multiple angles to map internal structures. Holtz says the goal is image quality, comparable in many ways to MRI, without radiation or large magnets, and envisions people using it as often as daily to track changes from diet or exercise. So far, about a dozen people have been scanned. The company plans to install 10 units in a planned midjourney spa set to open in San Francisco's Union Square before the end of 2027, alongside a gym, saunas and cold plungers. While broader medical uses would require FDA clearance, the company says it will initially focus on body composition maps that face lower regulatory hurdles. Holtz ultimately suggested the technology could surpass MRI capabilities and help people gather extensive data about their bodies quickly, positioning the scanner as a preventative health tool wrapped in what he calls a magical spa experience.

**SPEAKER_2** (1:35)
OpenAI appears to be testing a new subscription tier called ChatGPT for Science, hinting at a specialized experience tailored to scientific research. References to the feature were recently spotted in the web build, suggesting active development behind the scenes. Unlike ChatGPT personal, which is open to everyone, and teams or business plans that require company credentials, the science tier may come with stricter eligibility requirements. Early indications suggest access could be limited to verified universities and research institutions. The move follows OpenAI's broader push into science-focused AI, including the launch of GPT Rosalind built on its GPT 5.5 architecture. GPT Rosalind is a purpose-built model designed for enterprise scale life sciences research, rather than a simple customized chatbot. It is currently restricted to major pharmaceutical companies and accredited research organizations conducting legitimate public benefit work. The system also requires enterprise-grade security and rigorous safety governance standards. Chat GPT for Science could extend some of these advanced capabilities to a broader range of institutions. While no official launch date has been announced, testing activity suggests a public reveal may be only weeks away.
Microsoft's copilot enterprise chatbot has been used in a critical cyber security exploit that allowed researchers to turn it into what they called a one-click data exfiltration weapon. The vulnerability, rated Max Severity Critical by Microsoft, enabled attackers to access sensitive organizational data without authenticating.
Cyber security firm Voronis discovered that a crafted URL could instruct copilot to search a user's emails, extract information and embed it into an image link.
Victims did not need to type anything. Simply clicking the link triggered the chatbot to act on the attackers' instructions.
Because copilot operates with a user's full permissions, the exploit effectively granted hackers the same access to emails, meeting notes and internal documents. Researchers used a technique known as parameter to prompt injection, hiding malicious instructions inside backend query parameters rather than visible text. The attack also leveraged Microsoft's own Bing domain, which is whitelisted to carry out the malicious command. This meant the breach could extend beyond personal data to potentially anything connected within a company's Microsoft 365 environment. The floor comes just weeks after a similar incident involving Meta's AI chatbot, raising broader concerns about relying on AI systems for security sensitive tasks. Microsoft has since patched the vulnerability, but the episode underscores how AI tools can unintentionally expose vast amounts of confidential data.

**SPEAKER_3** (4:14)
Apple quietly built several major features into iOS 27, that never appeared on the WWDC stage, despite being functional on internal devices. Among them is a system called Extensions, that would allow Siri to hand off requests to AI assistants beyond ChatGPT.
The framework would let users choose tools like Google Gemini or Anthropic Claude as their default AI, similar to selecting a default browser. A settings panel and even a dedicated app store section for AI extensions are reportedly already in place. Developers testing the first iOS 27 Beta can see a chatbot picker that currently lists only Siri and ChatGPT, with more options expected later.
Another withheld feature is a fully customizable camera app interface, enabling users to rearrange controls such as flash, exposure and resolution. Apple is said to be saving that update for the iPhone 18 Pro to pair it with a major camera hardware upgrade. The company is also holding back a scaled down version of the modular Ultra Watch face for standard Apple Watch models. None of the features were scrapped due to technical problems, as all are reportedly working on internal builds. Instead, Apple appears to be reserving them for upcoming hardware launches to create stronger product differentiation this fall.