Midjourney Enters Medical Imaging — Full-Body Ultrasound Scanner Touted as "First New Modality in 50 Years"
Midjourney, best known for its image generation AI, has announced its entry into the medical imaging field. The company unveiled a full-body ultrasound CT system dubbed the "Midjourney Scanner." Founder David Holz described it as "a new whole-body medical imaging modality, the first in 50 years."
The scanner features roughly 358,000 ultrasound elements and, unlike MRI or X-ray, visualizes internal body structures without using radiation. It reportedly offers a resolution of about 0.5mm, fine enough to capture organ segmentation and tissue boundaries.
The system is currently at the prototype stage, and while AI is not yet used for image reconstruction, the company plans to introduce AI-assisted analysis in the future. Midjourney also plans to open "Midjourney Spa," a spa-integrated scanning facility near Union Square in San Francisco, in late 2027. The company is self-funded with no outside investment, and it has already begun discussions with the FDA.
Holz says he eventually wants to deploy 50,000 scanners and reach one billion scans per month. On Hacker News and elsewhere, however, some have raised questions about whether the technology is still at a preparatory stage and how practical it actually is.
Critical Vulnerability in Firefox's AI Features — Risk of Email Theft
A security researcher has disclosed a vulnerability that allows attackers to exploit Firefox's AI features to steal users' email contents. According to the vulnerability disclosure from insinuator.net, the AI capabilities built into Firefox can be leveraged as an attack vector.
As AI features in browsers expand, the case demonstrates that new attack surfaces are emerging in exchange for convenience. The detailed technical contents are available on the disclosure page. The security design of AI features will require ongoing scrutiny going forward.
73% of Japanese Companies Inadequately Prepared for "Shadow AI" — Gartner Survey
A survey by research firm Gartner has revealed that 73% of Japanese companies are failing to keep up with "shadow AI" — the phenomenon of employees using AI tools and services that the company has not officially sanctioned for work.
The traditional governance approach of "use only the AI the company has chosen" is no longer realistic. Designing a balance between employees' pursuit of convenience and security and compliance requirements has become an urgent challenge for enterprises.
Microsoft Integrates Its In-House Coding AI "MAI-Code-1-Flash" into Copilot
Attention is being drawn to the fact that GitHub Copilot in VS Code now offers a new model, "MAI-Code-1-Flash" — one that is neither GPT, Claude, nor Gemini. It is a coding-specialized model that Microsoft developed from scratch on its own.
Microsoft now offers a total of seven in-house AI models in its "MAI" series, all available in Copilot. The move appears to be a strategy to reduce dependence on OpenAI and differentiate the coding experience with its own models. Detailed performance evaluations remain to be seen, but this can be seen as part of a broader trend of major cloud providers building their own foundation models.
Google DeepMind's "Aletheia" Achieves Over 90% on IMO-Level Math Proofs
A write-up has been published reporting that Google DeepMind's mathematical proof system "Aletheia" achieved a success rate of over 90% on International Mathematical Olympiad (IMO)-level problems.
AI-driven mathematical reasoning has advanced rapidly in recent years, and Aletheia's results represent the frontier of that progress. What matters is that, unlike conventional pattern matching, the system can construct formal proof procedures — and expectations are rising for its future use as an assistance tool for mathematical research.