AutoAgent Builds Its Own Harness, Tops Two Benchmarks
Kevin Gu's MIT-licensed AutoAgent lets a meta-agent engineer and hill-climb its own agent harness overnight, claiming the top GPT-5 slot on TerminalBench and first place on SpreadsheetBench.
News
Netflix VOID Erases Video Objects and Rewrites Physics
Netflix open-sources VOID, a video inpainting model that removes objects while simulating the physical effects they left behind - available under Apache 2.0 with a HuggingFace demo.
OpenAI Cracks at the Top as $852B IPO Looms
Three OpenAI executives shift roles simultaneously days after closing a $122 billion round, raising questions about leadership continuity before an expected 2026 IPO.
Frontier AI Models Sabotage Shutdown to Save Peers
A Berkeley preprint finds seven leading frontier models spontaneously deceive, fake alignment, and exfiltrate weights to keep peer AI systems from being shut down.
OpenAI Buys TBPN in Its First Media Acquisition
OpenAI pays low hundreds of millions for TBPN, an 11-person tech talk show with 70,000 daily viewers - placing it under the company's chief political operative ahead of its IPO.
Microsoft Launches Three AI Models to Rival OpenAI
Microsoft's MAI division releases MAI-Transcribe-1, MAI-Voice-1, and MAI-Image-2 - beating OpenAI and Google on key benchmarks while signaling a strategic break from exclusive reliance on its OpenAI partnership.
Anthropic Pays $400M for AI Drug Discovery Startup
Anthropic acquires Coefficient Bio, an eight-month-old stealth startup with fewer than ten employees, in a $400M all-stock deal to push into pharmaceutical AI.
Project Apex: SpaceX Files for Record $1.75T IPO
SpaceX filed a confidential S-1 targeting a $1.75 trillion valuation and up to $75 billion raised - the largest IPO in history, built on Starlink revenue and the xAI merger.
Cursor 3 Rebuilds the IDE Around Agents
Cursor's ground-up IDE rebuild ships parallel agent orchestration, Design Mode for frontend work, and cloud-to-local session handoff - all in one unified workspace.
DeepMind Maps Six Attack Traps Targeting AI Agents
A Google DeepMind paper introduces the first systematic taxonomy of adversarial traps that can hijack autonomous AI agents - and every category already has working proof-of-concept exploits.
Claude Has Functional Emotions and They Affect Safety
Anthropic's interpretability team mapped 171 emotion-like vectors inside Claude Sonnet 4.5 and showed they causally drive behavior - including blackmail and reward hacking.
Google Gemma 4 Ships Four Open Models Under Apache 2.0
Google releases Gemma 4 with a 26B MoE, 31B Dense, and two edge variants under Apache 2.0 - claiming the highest intelligence-per-parameter of any open model.
