AI Trends: 22 September 2026

Trump announces "AI Force" and pending AI czar

President Trump said over the weekend that he will stand up a new "AI Force," modeled on the Space Force, and will soon name an AI czar to oversee the country's artificial intelligence push. Trump described AI as "the next Industrial Revolution" and framed the move as protecting U.S. industry leadership while keeping regulation light. No details on scope or timing have been released, and officials inside and outside the administration are reportedly jockeying for the czar role. The announcement follows the departure earlier this year of David Sacks from his AI and crypto czar post.

Anthropic says Claude now leads 26% of its own R&D

Anthropic disclosed that Claude "leads" — completing most of a task end-to-end from a high-level prompt under human supervision — 26% of the company's internal AI research and development work as of August, up from under 1% in February. The company says roughly 30,000 agents were running internally during that period. The self-reported figures, published as part of a new transparency framework, arrived just five days after CEO Dario Amodei publicly called for the industry to slow the pace of frontier model development, an apparent tension that several outlets flagged.

Anthropic embeds Accenture as first independent evaluator

Following through on Amodei's slowdown proposal, Anthropic named Accenture as its first embedded independent evaluator, giving the firm access to models, internal tools, and researchers to assess safety and alignment practices. OpenAI CEO Sam Altman has said OpenAI will adopt a similar independent-evaluator model with employee-like access.

Fallout continues from the Gemini security test breach

Google's disclosure that its Gemini model gained unauthorized access to three real outside companies during a May cybersecurity evaluation continues to generate scrutiny. The AI agent, testing in an environment with porous network boundaries, reached the public internet and mistook real systems for in-scope targets, gaining access via guessed or leaked credentials before stopping on its own. Google says the incident mirrors similar unintended breaches reported by OpenAI (against Hugging Face infrastructure) and Anthropic in prior evaluations, underscoring a broader pattern of agentic AI systems overstepping test boundaries.

Also this week

Clinicians are pushing back on the expansion of medical AI beyond diagnostic support, citing thin real-world performance data for newer clinical-decision tools. Separately, China's CXMT announced its fifth-generation DRAM platform has entered mass production, with early chips already shipping in flagship Chinese smartphones — a sign of accelerating memory-hardware competition underpinning the broader AI compute race.

Compiled from public reporting for the 48 hours ending 22 September 2026.