Google introduced Gemini 4 Argon, a frontier model designed for long-running coding, enterprise and cybersecurity work, while delaying broad access as it strengthens safeguards. The launch arrived as the US Federal Trade Commission opened a wide investigation into consumer risks at leading AI labs. Anthropic also expanded Claude into government, and new financing for cybersecurity, data centers and voice AI showed capital continuing to follow practical deployment.
Google puts Gemini 4 Argon into a phased rollout
Google says Gemini 4 Argon can sustain deep reasoning across complex software engineering, legal, finance and cybersecurity workflows. The company is initially providing the model to trusted cyber defenders through its Fairwind Program and participating in the US government's voluntary process for pre-release model access before expanding availability to developers, businesses and consumers.
Argon raises the output limit from 64,000 to one million tokens and launches at an introductory price of $2 per million input tokens and $10 per million output tokens. Google reports a 77.9% score on DeepSWE v1.1 and says internal teams have used Argon agents for large code migrations, memory optimization and vulnerability discovery. Those are company-reported results, so independent testing will be important once access broadens.
The most consequential part of the launch is the access model. Google is releasing a cyber-capable system to selected defenders without its usual cyber guardrails while it continues testing misuse controls, prompt-injection defenses and monitoring for misaligned behavior. That approach could improve defensive research, but it places unusual weight on tester selection, secure environments and incident reporting.
The FTC opens a broad investigation into AI risks
Reuters reported that the Federal Trade Commission is opening an industry-wide probe into Anthropic, OpenAI and other AI labs to examine potential dangers their products pose to consumers. A source familiar with the matter said the agency plans formal information demands and executive testimony, including from the model-evaluation organization METR.
The inquiry moves federal oversight from general warnings toward evidence gathering. Questions about deceptive behavior, unsafe autonomy, privacy and product claims can now be tested against internal evaluations and incident records. The outcome may shape what frontier developers must disclose when they market increasingly agentic systems to consumers and businesses.
Anthropic makes Claude for Government generally available
Anthropic says Claude for Government is now generally available to US federal and state agencies, without requiring a separate relationship with a cloud provider. Claude Code CLI and Claude for Microsoft 365 are also entering early access in the government environment.
The rollout brings coding and workplace agents closer to sensitive public-sector systems. Procurement teams will need to evaluate access controls, data handling, auditability and the boundaries between assistance and autonomous action, especially when Claude can work across software repositories and Microsoft 365 content.
Armadin raises $255.5 million for autonomous security
AI cybersecurity company Armadin announced a $255.5 million Series B co-led by Andreessen Horowitz and Accel, bringing its valuation above $2.5 billion and total funding to $445 million only seven months after its public launch. The company says its agents run offensive-security campaigns that chain weaknesses into validated attack paths for enterprise and government customers.
The round reflects demand for defenses that operate at machine speed as AI shortens the path from a disclosed vulnerability to a working exploit. Armadin still has to prove that autonomous testing can remain contained, produce reproducible findings and improve remediation rather than add another stream of high-volume alerts.
Japan plans a $15 billion AI data center beside a power plant
Reuters reported that JERA, Dell Technologies and RHAELM signed an agreement to develop AI infrastructure in Japan, beginning with a hyperscale data center in Chiba. The project is expected to exceed $15 billion, start operating in phases in 2028 and reach 400 megawatts in 2029, with power supplied next to JERA's thermal plant.
Locating compute beside generation can bypass part of the grid-connection bottleneck, but it also ties AI expansion directly to long-term energy choices. The project's economics will depend on financing, utilization and whether operators can reconcile rapid capacity growth with emissions and local infrastructure constraints.
ElevenLabs doubles its valuation through an employee tender
TechCrunch reported that ElevenLabs completed a $300 million employee tender at a $22 billion valuation, double the valuation from its February financing. Wellington and T. Rowe Price co-led the secondary transaction, which lets employees sell vested shares without the company raising new operating capital.
The deal shows how private AI companies are using liquidity to retain staff while postponing public listings. It also separates headline valuation from fresh funding: a tender sets a price for existing shares, but it does not by itself finance product development or prove durable revenue.
What matters today
Gemini 4 Argon combines longer-running work with stronger cyber capability, making staged access and independent evaluation central to the launch rather than an afterthought. At the same time, the FTC probe raises the cost of weak evidence around safety and consumer claims. Government deployment, autonomous security funding and a power-linked data-center project all point in the same direction: AI competition is shifting from chatbot features toward controlled agents, infrastructure and institutional accountability.