U.S. Gives Government Early Voluntary Access to Powerful AI Models

Overview

• The order, signed Tuesday, lets Treasury, NSA and CISA create a coordination center and request early access to advanced models for a voluntary 30-day security review before public release.
• News leaders, led by The New York Times' Arthur Gregg Sulzberger at a global editors’ meeting, say major AI firms have used newsroom content without permission and are pressing for transparency, licensing and rules to protect journalism revenue and provenance.
• Security analyses show a surge in AI-enabled attacks in 2025, with reports of more realistic deepfakes, automated phishing and data-theft campaigns that make continuous monitoring and industry-government cooperation more urgent.
• Industry research and consultants report most corporate AI pilots fail to move the needle on profit because projects remain bolt-on efforts instead of reorganizing data, governance and operations; a small set of firms that redesign around AI have reported large productivity gains.
• Vendors and developers are reacting to the new review regime in different ways, from Anthropic expanding selective Mythos access for security partners to NVIDIA pitching local PC chips to run agents, a mix that highlights tensions between faster deployment, national security and company control.