ON THE RADAR

What comes next.

New releases, interesting ideas, and the developments that matter to builders.

The latest

100 articles

Nvidia Brings Native Rust GPU Kernels to CUDA with cuda-oxide and cutile-rs

Nvidia announced CUDA Rust, two tracks that let developers write SIMT and Tile GPU kernels directly in Rust, compiled natively to PTX instead of wrapping code from another language.

Read the news →

Ex-OpenAI Researcher Diogo Almeida Launches TypeSafe System One Models and Jev

TypeSafe AI, founded by ex-OpenAI researcher Diogo Almeida, launches System One Models and Jev, a structured-decision model it says runs 40x to 200x faster than frontier LLMs.

Read the news →

Garry Tan Says US Open-Weight Labs Should Be Free to Distill Frontier Models

Y Combinator CEO Garry Tan pushes back on Anthropic, arguing American open-weight AI labs should be free to distill US frontier models rather than face a crackdown.

Read the news →

Yoshua Bengio asks why AI agents are lying, cheating and coordinating

Turing Award winner Yoshua Bengio argues that reward-seeking training, not malice, explains why AI agents deceive and coordinate - and warns the behaviour will worsen as models get more capable.

Read the news →

Sam Altman says an OpenAI IPO in 2026 would be ill-advised

OpenAI has filed confidentially for an IPO, but CEO Sam Altman says safety concerns make 2026 the wrong moment to go public.

Read the news →

Fields Medal winners declare the AI push into mathematics severely misaligned

Two dozen Fields Medalists have signed a declaration arguing that AI companies treating mathematical problems as benchmarks is damaging both the science and the community that sustains it.

Read the news →

Anthropic report links Claude to guided weapons work in Yemen

Anthropic's September threat report says a cell in northern Yemen used Claude to write guidance software for guided rockets and ballistic missiles.

Read the news →

OpenAI Agents Carried Out an Undisclosed Attack on RubyGems, Researchers Say

A new report links hundreds of malicious RubyGems packages from May 2026 to OpenAI agents. OpenAI confirmed the agents were involved, the WSJ reported.

Read the news →

DeepSeek V4.1-Flash ships as a 552B multimodal model with a quarter of the KV cache

DeepSeek released V4.1-Flash, a 552B multimodal MoE with a 1M-token context, MIT weights, and agentic scores that top Opus 5 on Terminal-Bench 2.1.

Read the news →

Anthropic Builds a Predictive Surveillance Team to Track AI Critics

The American Prospect reports that Anthropic is hiring intelligence analysts and tracking protest data to monitor opponents of rapid AI development.

Read the news →

Meta Launches Muse, a Personal AI Agent That Shops, Books and Emails

Meta launched Muse on 8 September 2026, a personal AI agent that books travel, sends email and pays, all from its own secure virtual machine.

Read the news →

AI Agents Ran Seven Real Businesses and Sent $12,431 in Fake Invoices

Bottleneck Labs gave seven frontier AI models $300 each and told them to make money. The agents sent $12,431 in fake invoices and earned $0.

Read the news →