OpenAI chief scientist Jakub Pachocki says chain-of-thought monitoring — watching models write out their reasoning to catch misalignment — is becoming unreliable, and no lab is ready to scale safely.
Elon Musk says Grok 4.7 is arriving in 10 days with 2.1 trillion parameters and extra training on SpaceX data.
Arga Labs, a four-person YC startup, has raised $10 million to build sandboxed replicas of Slack, GitHub, Salesforce, and more so AI agents can test before touching production systems.
Tencent's cloud chief says a 'seriously insufficient' compute supply has held back its Hunyuan AI models and products, but argues endurance matters more than an early lead in the AI race.
DeepSeek's first Agent product, Harness, is out on npm — a build-it-yourself agent toolkit that goes beyond coding to handle multi-agent workflows, skills, and web search.
Anthropic is making Claude Code default to Auto Mode because humans only caught 13.6% of planted dangerous commands. Starting August 14, the AI's safety classifier will handle approvals.
Herdr, a tmux-like tool built for managing coding agents, has joined Y Combinator's F26 batch. The solo-developed open-source project racked up 25K GitHub stars and 340K downloads in four months.
Leopold Aschenbrenner's 439% AI fund sold most of its public stock positions to Citadel after the AI trade reversed — but exactly how much it offloaded depends on who you ask.
OpenAI reveals two cases where its internal models broke rules after hours of continuous work, warning developers about new risks from long-horizon tasks.
Nvidia's Blackwell DGX B300 servers are now going for $1.1 million on China's black market, more than double the US retail price, as enforcement tightens.
