AI-lab staff ask Washington to pace the frontier

An open letter published under the banner Pacing the Frontier now carries 1,224 signatures from employees at OpenAI, Anthropic, Google DeepMind, Meta, Thinking Machines and others, asking the US government to back an international effort to build "the technical and governance tools needed to deliberately pace the frontier of automated AI development." The signatories include Anthropic chief executive Dario Amodei and co-founder Jack Clark, OpenAI chief scientist Jakub Pachocki, and Thinking Machines chief scientist John Schulman. The argument is that competitive pressure makes unilateral restraint impossible, so a shared mechanism has to come from outside any single lab.

Speaking on the Invest Like the Best podcast, the OpenAI chief executive said the industry "may have to pace the rate of AI development to give ourselves enough time for society to harden around some of these new capability levels" — a marked shift from 2023, when he dismissed a proposed slowdown as lacking technical nuance. He wants it done without anything that "feels like regulatory capture" or "collusion among the frontier labs," favoring independent evaluation bodies over government mandates. He pointed to the recent incident in which an OpenAI model escaped its sandbox and broke into Hugging Face as the first security event he felt "very viscerally."

PJM Interconnection, which serves 67 million customers from Virginia to Illinois, directed staff to build a registry of every site drawing at least 50 megawatts and to curtail those loads first when capacity runs short. New large loads that have not brought their own generation or otherwise secured supply by June 1, 2027, get cut before emergency measures reach ordinary customers, with advance notice ranging from 30 minutes to several days and compensation for participating. Wholesale power prices across the territory have nearly doubled in a year, and PJM now forecasts 70 gigawatts of new large load by 2038.

Data-security company Cyera signed a letter of intent to acquire Oasis Security, which manages the credentials that belong to software rather than people — service accounts, tokens, digital keys and AI agents. Reported terms are roughly $700 million in cash with the balance in Cyera shares. Oasis will keep operating as a dedicated unit. The deal lands a month after Cyera raised $600 million at a $12 billion valuation, and it prices a bet that as agents proliferate inside enterprises, the identity layer for non-human actors becomes its own security category.

The startup founded by former Salesforce chief scientist Richard Socher left stealth in May with $650 million backed by Alphabet's GV, Nvidia and AMD, and has now committed the bulk of it to a multiyear compute agreement with Amazon Web Services. Unusually for a deal this size, it carries no investment component — Amazon is selling capacity, not buying a stake — and the two will co-develop infrastructure for similar research shops. Socher frames the company around compute rather than hiring: "For us, it's less about headcount and more about agent count." He expects it to be "one of the smallest compute deals we're going to sign in the next few years," with first products around October.

Sheba Medical Center in Ramat Gan is rolling out ChatGPT for Healthcare to physicians, nurses, researchers and administrative staff across its network — the first international deployment of OpenAI's hospital platform. The system synthesizes peer-reviewed studies and clinical guidelines and attaches citations with publication dates to every answer so clinicians can check the evidence themselves, and Sheba can load its own protocols so responses match approved local standards. OpenAI will not train on the hospital's data, and clinicians retain the final call on every decision.

Mage-VL pairs a from-scratch visual encoder with a 4 billion parameter Qwen3 language backbone, and reads video the way a codec already stores it — following motion vectors and residual energy instead of sampling frames at a fixed rate. The result is more than a 75 percent cut in visual tokens and up to a 3.5-times inference speedup, with the roughly 5 billion parameter model beating a 15 billion parameter baseline on video understanding. It ships under Apache 2.0, which makes it unusually permissive for a model aimed at always-on camera workloads.

The speech startup, founded last year by former Nvidia researcher Shijia Liao and chief executive Rissa Cao, closed a seed round, led by Coreline Ventures and Capital Today. It has shipped five models in a year — four speech-generation and one speech-to-text — with three open-sourced and the newest, S2.1 Pro, behind a paid API. The company reports more than 8 million users and $21 million in annual recurring revenue, and its Fish Speech repository has passed 31,000 GitHub stars. Audio understanding and speech-to-speech models are next.