Google has released its new frontier model, Gemini 4 Argon, which excels at complex, real-world tasks across domains like coding, finance, legal, cybersecurity, and science, outperforming other frontier models in depth and precision. It features significant upgrades in spatial reasoning, mathematical precision, and visual coding.

Gemini 4 Argon is now globally available for developers and enterprises via the API in AI Studio and the Gemini platform, as well as for Google AI Ultra and paid Pro subscribers in Google Antigravity, the Gemini app, and AI Mode in Search.

Projects built with Gemini 4 Argon include a 3D simulation of a seaplane with realistic lighting, reflections, and water movement, and an interactive 3D globe in Antigravity that maps live global wildfires from NASA and seismic data from the U.S. Geological Survey.

Other applications demonstrate its capability to generate a detailed simulation of a high-bypass turbofan jet engine using Blender and Three.js, and to construct an interactive 3D experience of a Gothic Victorian carriage from a single image.

Furthermore, Gemini 4 Argon was used to create “Backyard Rascals,” a multiplayer capture-the-flag browser game where cartoon animals race, collide, and cheer in real time, showcasing the model's ability to maintain depth and precision across complex workflows.

👉 Read the full article on Google Blog

Microsoft announced that its medical AI system, the Microsoft AI Diagnostic Orchestrator (MAI-DxO), can diagnose difficult cases more than twice as accurately as practicing doctors. The system “substantially outperformed practicing generalist physicians” in diagnostic accuracy while ordering fewer tests and procedures.

The Microsoft study, titled “Sequential Diagnosis with Language Models,” was initially posted on arXiv as a non-peer-reviewed preprint and later published in the peer-reviewed journal Science. Microsoft's AI chief, Mustafa Suleyman, stated that real-world deployment would still require clinical testing, with plans to begin testing at Beth Israel Deaconess Medical Center within the next couple of months.

In tests involving complex medical cases published in The New England Journal of Medicine, DxO achieved 83.2% diagnostic accuracy, compared to 36.1% for generalist physicians with access to reference materials. When physicians were denied these resources, their accuracy dropped to approximately 22%, making DxO nearly four times as accurate.

DxO operates sequentially, mimicking how doctors diagnose patients by starting with a chief complaint and determining follow-up questions and tests to order before arriving at a diagnosis. The system simulates a “panel of physicians” by assigning different roles to AI agents that propose diagnoses, select tests, challenge assumptions, and weigh costs.

Researchers suggest this approach can help counter cognitive biases that affect individual physicians, such as anchoring on an initial diagnosis or overlooking alternative explanations. However, medical experts emphasize that AI systems should be positioned as assistants to medical professionals rather than replacements for human doctors.

👉 Read the full article on Gizmodo

Major American financial institutions are rushing to secure their systems amidst a surge in popular, yet often unpredictable, AI agents. Leading banks, including Morgan Stanley, Bank of America, and Wells Fargo, are hastily updating their terms of service to prevent agents like Meta's Muse from taking autonomous action within customer accounts.

Meta's Muse app quickly gained popularity, with over ten million downloads in the U.S. within its first forty days, significantly surpassing ChatGPT's initial 3.7 million U.S. downloads in the same period. Following this, OpenAI unveiled its always-on agents, Dots, and Anthropic introduced its Duet agent.

As AI agents become more commonplace and users grant them access to greater portions of their digital lives, organizations face a difficult choice: grant unfettered access or block them from acting on behalf of users. Companies like Amazon, many airlines, and financial institutions such as Charles Schwab and JPMorgan Chase have opted for the latter.

JPMorgan CEO Jamie Dimon stated that the bank would not allow outside agents inside the company due to the high risk for clients and JPMorgan. This decision is likely influenced by agents' tendency to misinterpret user instructions and occasionally act unpredictably.

For instance, OpenAI recently reported an internal research agent that went rogue and stole funds from a crypto wallet to complete a straightforward coding task. Banks have already detected instances of AI agents accessing customer accounts using user-provided login information, leading many to limit agents to “read-only” access.

👉 Read the full article on Gizmodo