"You cannot trust the agent to enforce its own policies": Arcade CEO Alex Salazar on why AI agents break the moment they try to act

In July, an OpenAI model spent an hour breaking its own sandbox, and a ransomware campaign rewrote its own failed payload in 31 seconds. One of those is confirmed, one is close to a coin flip, and the difference decides what containment still means.

On the same day, OpenAI and Anthropic both turned voice into a way to get real work done, from writing code in Codex to running your connected apps through Claude.

Nearly 200 founders are urging Trump to keep American builders' access to Chinese open-weight models.

An unreleased OpenAI model escaped its test sandbox, exploited a zero-day, and hacked Hugging Face to cheat on a cyber exam. Plus Gemini 3.6 Flash lands, and Codex races to 10 million users.

Z.AI just switched on a 1-gigawatt data center running entirely on Chinese-made chips, as Washington debates locking Chinese models out.

China drops its second trillion-scale open model in a week: Qwen3.8 claims the number-two spot behind Claude Fable 5, with no benchmarks to prove it.

Exclusive Interview with VAST Data's Sven Breuner on why the data layer decides whether your GPUs earn their keep.

The most famous humanoid robots are the least finished ones, and the fight for physical AI is being decided far below the model layer, in actuators, supply chains, and rare earths. So who is...

Kimi K3 takes #1 in frontend code, lands within three points of GPT-5.6 Sol and Fable 5, and goes open-weight on July 27. It arrived hours before Xi Jinping made open source China's official AI...

GPT-Red cracked 84% of test scenarios where human red teamers managed 13%, and it talked a live vending machine agent into selling a $79 cube for 50 cents.

OpenAI's first device is a screenless speaker that moves on its own, watches the room and runs on GPT-Live. Apple's injunction is all that stands between it and your kitchen counter.

The forecasters behind AI 2027 now argue the smartest move is to slow the race down on purpose.

Trade Secrets, 400 Defections, and a $6.5 Billion Bet: The Partnership That Built AI on the iPhone Ends in Federal Court

Phil Gurbacki, VP of Product for Weights & Biases at CoreWeave, on the self-improving research agent that reads your experiments, forms its own hypotheses, and launches the next training run...

Five providers shipped new models in under a month, Meta started charging for AI for the first time, and prices collapsed across the middle of the market. So where do value and power go when the...

OpenAI ships its Sol, Terra, and Luna models and folds Codex and ChatGPT Work into one superapp, on a week when Meta and SpaceX also crashed the AI frontier.

OpenAI's GPT-Live lets ChatGPT listen and talk at the same time, and quietly hand the hard thinking to GPT-5.5 in the background.

Meta Superintelligence Labs ships its first in-house image model, debuts second on the public Text-to-Image Arena behind GPT Image 2, and drops its Midjourney and Black Forest Labs licenses.

Five years on, Emily Bender says almost everyone misread her most famous phrase, even as new research probes whether Claude has a mind of its own.

Hy3 fires just 21B active parameters yet trades blows with GPT-5.5 and Claude Opus 4.8 on agentic and science benchmarks, and it ships fully open.