Anthropic Announces 'Claude Corps' — a $150M Program to Train 1,000 Fellows
On June 11, Anthropic announced Claude Corps, a nationwide fellowship program designed to bring the benefits of AI to local communities. The program will recruit 1,000 early-career professionals, give them intensive training in how to use Claude, and then place them full-time for a year at nonprofit organizations across the United States.
The program rests on a partnership among three organizations: Anthropic provides the funding and overall strategy, CodePath (the largest university computer science education nonprofit in the US) serves as the fellows' employer and leads the training, and Social Finance handles impact measurement and the design of financing schemes for future scaling.
The initial investment is $150 million. Anthropic stated that "the benefits of transformative AI systems may come at the cost of significant disruption," emphasizing that the companies building the technology have a responsibility to invest directly in workers.
Why it matters: This stands out as a concrete example of an AI company putting real investment behind its response to the technology's impact on jobs. If Claude Corps delivers results, it could become a model for extending AI's benefits during a period of economic transformation — and the foundation for a much larger rollout.
Google DeepMind Invests $10M in Multi-Agent AI Safety Research
Google DeepMind, together with Schmidt Sciences, the UK's Advanced Research + Invention Agency (ARIA), the Cooperative AI Foundation, and Google.org, has announced up to $10 million in funding for research on the safety of multi-agent AI systems.
The backdrop is the arrival of a "multi-agent era" in which millions of AI agents built by different organizations communicate and transact with one another in digital environments. Today's safety evaluations mostly test individual models in isolation, but when agents interact with one another, there is a risk of unexpected collective (emergent) behavior.
According to MIT Technology Review, Rohin Shah, who leads AGI safety and alignment research at DeepMind, warned that a proliferation of agents that carry out tasks without human oversight and follow instructions from other agents creates "a whole new class of risks." The funding is intended to advance academic research outside of industry labs.
Why it matters: As AI agents move rapidly into production use, the focus of safety research is shifting from the safety of individual models to the safety of interactions between systems. With warnings mounting about potential surges of unpredictable economic activity and new security challenges, up-front investment in this research matters all the more.
AI Boom Is Fueling Inflation Through Soaring Memory Chip Prices, Bloomberg Reports
According to a June 11 Bloomberg report, the surge in memory chip prices driven by AI demand has reached "insane" levels and has become one of the factors pushing up US inflation. Demand for HBM (high-bandwidth memory) and DRAM from AI data centers continues to far outstrip supply, and rising chip prices are starting to feed through into consumer prices.
Why it matters: This suggests the economic side effects of AI investment are beginning to show up in macroeconomic indicators. The AI boom is affecting not just the tech industry but the entire semiconductor supply chain — and ultimately the prices everyday consumers pay.
Trend Micro Warns: AI Agents Are Already Compromised
Trend Micro has published the first installment of its "Pwning Agentic AI" series, a report titled "Your AI Agent Is Already Compromised." It offers a detailed analysis of just how vulnerable AI agents are to external attacks.
The report also drew attention on Hacker News, underscoring that AI agent security has become an urgent issue in production environments.
Relatedly, reports have also been making the rounds that an AI agent deployed in Verizon's billing department caused serious problems in customer interactions. While AI agents are seeing real-world adoption, cases where the gap between expectations and the actual customer experience becomes glaring are on the rise.
Why it matters: As AI agents spread, security vulnerabilities and operational challenges are surfacing at the same time. The more autonomously an agent acts, the larger its attack surface grows. Running agents safely demands an approach different from traditional application security.
Ex-Hugging Face Pretraining Team Releases 'Refiner,' a Data Library for Robotics
Macrodata Labs, founded by members of Hugging Face's former pretraining team, has released Refiner, an open-source library built specifically for refining and processing robotics data.
It supports ingesting all the major robotics data formats, including Parquet, HDF5, MCAP, Zarr, RLDS, and LeRobot, and provides integrated support for processing steps common in robotics ML pipelines, such as visual hand tracking, subtask annotation, and running reward models.
Why it matters: A proliferation of incompatible data formats has long been a pain point in robotics. Refiner could play the same role in robotics that Hugging Face played in NLP — unifying the ecosystem around shared data and model hubs.
Offsend — A Local Mac App That Keeps Secrets Away from AI Tools
Offsend, an open-source project, has released a Mac app that runs locally and masks sensitive information before it ever reaches AI tools. It automatically replaces secrets such as API keys and passwords before an AI coding assistant can read them.
Why it matters: It's drawing attention as a practical approach to preventing secret leakage when using AI coding tools. Demand is expected to grow for solutions like this, which strike a balance between security and convenience when rolling out AI tools in enterprise environments.
---