AI Weekly

Tapes Together Strong: The Co-evolution of Computation and Cooperation

The essay opens flat. "Across this four-part series, we have argued that Europe's AI ecosystem is subject to the gravitational pull of dominant US players," write Frederike Kaltheuner and Leevi Saari of the AI Now Institute in

Coupling computational capacity to a shared energy budget can make cooperation self-organize, according to a preprint titled

Italian embedded-security firm Exein raised $270M led by Headline at a $1.7B post-money valuation, with Sofina, Goldman Sachs, EIB Group ETCI and KfW Capital co-investing. Its kernel-level Photon runtime already ships in over 2 billion chips at customers including Nvidia, Arm, Mediatek and AWS; first-half 2026 ARR grew four-fold. Exein plans an agentic security architecture by end of 2026 and a foundation model for tracking and securing physical AI systems in early 2027.

Bloomberg reports Kioxia is exploring a US ADR listing to raise at least $10B in spring 2027, working with Bank of America, Goldman Sachs and JPMorgan. The Tokyo-based memory maker wants US liquidity after billions of dollars in Japan share buybacks and eligibility for semiconductor-focused stock indices. A listing that size would be one of the largest AI-memory float events amid HBM demand from Nvidia and hyperscalers. Details, size and bank lineup remain preliminary.

A new arXiv paper introduces RSIAgent, a training-free multi-agent framework using curriculum, actor and verifier agents to build reusable environment memory without touching model weights. Its broad-then-deep exploration strategy captures causal action-outcome traces and hidden constraints. On OSWorld-v2 and Agent's Last Exam, RSIAgent lets open-source Kimi K3 and GLM-5.3 outperform frontier closed-source systems including GPT-6, and the frozen memory transfers to downstream tasks without fine-tuning.

Shanghai AI Laboratory released Atria Dawn Preview, a 744B-parameter agentic MoE built on GLM-5.2 and trained via a Verifiable Experience Pipeline that grounds tool use in executable environments. The paper reports competitive results across 16 benchmarks with top scores on five, drawing on 769 task records from 56 participants; roughly one-third of AI-assisted tasks were judged infeasible without the agent. Signed by 143 authors, it frames RSI-friendly R&D as project-level human-AI partnership rather than task execution.

A special edition drawn from what the experts we follow shared and discussed last week, August 31–September 6.

Jacob Coxon left Anthropic and accused the leading AI labs of moving faster than they can make their systems safe. His warning lands as researchers still inside those labs say the same core problem remains open.

Dario Amodei, chief executive of Anthropic, the company behind the Claude AI assistant, is calling for slower development of the most powerful AI systems. OpenAI chief executive Sam Altman and xAI founder Elon Musk have backed him. They call this “pacing”: allowing more time for safety work as AI improves.