Will Anthropic Keep Supplying Claude to the NSA? What the "Supply Chain Risk" Designation Means
According to a May 24 report by The Decoder, the NSA (National Security Agency) will likely continue to be supplied with Claude models, despite the U.S. Department of Defense flagging Anthropic as a "supply chain risk" company.
The backdrop: U.S. intelligence agencies don't have enough of Nvidia's latest Grace Blackwell chips. Anthropic's "Mythos" model is reportedly able to run on older-generation hardware, making it a valuable option for intelligence agencies facing chip shortages. Notably, the controversial "any lawful use" clause that stalled earlier negotiations is reportedly not part of this arrangement.
At the same time, debate is expected to continue over how national security concerns can coexist with partnerships involving AI companies, and over transparency around surveillance.
Claude Code Discovers a Scaling Algorithm No Human Designed
A research team from the University of Maryland, Google, Meta, and others has introduced "AutoTTS," a system that uses Claude Code as a coding agent to autonomously discover control algorithms for AI inference.
The algorithm Claude Code found in this study reportedly maintains accuracy on par with the standard self-consistency method while cutting compute by roughly 70%. What's especially noteworthy: the search cost just $40 and took only 160 minutes.
The fact that AI itself discovered an approach no human researcher had designed makes this a significant result, demonstrating the potential of AI-automated AI research (AI-for-AI).
California Signs Measures to Address AI Job Displacement
California Governor Newsom has signed an executive order aimed at dealing with job losses caused by AI, The Hill reported on May 24.
The order is intended to strengthen worker protections and retraining as AI adoption rapidly transforms the labor market. As home to much of the tech industry, California could serve as a test case for AI job displacement policy, offering lessons for other states and the federal government.
Former Tencent AI Lead: "China Is Behind in the LLM Race, but Can Win in AI"
In an interview with the South China Morning Post (SCMP), a former head of Tencent's AI division offered an analysis of the state of China's LLM development.
According to him, China currently trails the U.S. in LLM performance, but it can gain the upper hand in applied AI. Details on specific strategies and domains are limited, but he argued that the race for LLM performance is not everything when it comes to AI dominance.
AI Is Destabilizing the Labor Market — Karen Hao Warns
Journalist Karen Hao has published an analysis warning that AI is creating a "desperate class of workers" without full-time employment.
She points out that as AI tools spread, companies are accelerating their shift from full-time employment toward freelance and contract-based work, eroding workers' economic stability. The theme dovetails with policy developments in California, and debate looks set to expand as a key issue in AI's social impact.
---