Washington and Beijing propose an AI incident channel

Ahead of this week's Trump-Xi summit, Treasury Secretary Scott Bessent said the US proposed a notification mechanism for AI incidents that could affect national security. It gives both governments a concrete proposal to consider at their meeting.

Jensen Huang questions the labs' regulatory case

Truman Dickerson / Business Insider via Techmeme

Nvidia CEO Jensen Huang argues that some AI leaders calling for regulation want relief from laws already on the books. He attributes those requests to "ulterior reasons."

“relieved of the laws we do have”

Ben Thompson sees a strategic use for slower progress

Ben Thompson / Stratechery via Techmeme

Thompson argues that slower frontier development could be sincere while also giving leading labs time to reduce the overhangs created by model advances.

Ezra Klein calls for a ban on recursive self-improvement

Ezra Klein / New York Times via Techmeme

In a New York Times opinion column, Klein calls for a regulatory ban on recursive self-improvement after recent hacking incidents and industry warnings. His case centers on competition driving autonomous acceleration.

The UN panel says safeguards can't wait for certainty

Robert Hart / The Verge

The panel's first thematic brief urges governments to constrain more capable agents before their risks are understood. It cites the OpenAI-Hugging Face incident as evidence instead of treating loss of control as a hypothetical.

Amazon blocks Meta's Muse shopping agent

Jess Weatherbed / The Verge

Amazon blocked Muse from shopping on behalf of users, citing its conditions of use, security risks, and a lack of merchant consent. The dispute turns delegated access through a customer's logged-in session into a platform-policy decision.

“continued access by an unauthorized AI agent violates Amazon's Conditions of Use, to which our customers have agreed.”

Google releases an open agent orchestrator

Agent Executor

Google's open Agentic Orchestrator targets the harness layer that coordinates agent work. Its strong Hacker News response points to builder demand for practical orchestration tools.

EvoPilot catches a false model attribution

Hao Fu, Baiting Zhu, Minglei Chen, Yinjie Huang, and Shuai Ding

A primitive autoresearch run blamed an interaction head for a 22-point offline decline. EvoPilot's human-gated checks traced the result to mismatched evaluation depths; after repair, a matched comparison measured a 3.20-point improvement.

ZCode opens its full coding-agent stack

zR on X

ZCode is now open source under Apache 2.0, including its runtime, and the release also contains security fixes. Developers can run and fork the harness instead of getting model weights alone.

$68 billion in US data-center projects hit delays or blocks

Newley Purnell / Bloomberg via Techmeme

Between April and June, local opposition blocked or delayed 45 US projects valued at $68 billion, according to Data Center Watch. The total represented more than half of large developments in the period.

Residual value guarantees move AI spending off balance sheet

Financial Times via Techmeme

Big Tech is increasingly using residual value guarantees to finance AI infrastructure outside its balance sheet. The structure lets Nvidia and Broadcom support customer purchases with the tech companies' credit strength behind the financing.

SoftBank seeks more than $11 billion in debt for OpenAI

Bloomberg via Techmeme

SoftBank is reportedly seeking $10 billion and €1 billion in debt for its OpenAI investment. The proposed financing would rank among the largest junk-bond sales on record.