More than 1,100 researchers and engineers from leading AI labs signed a public statement on Tuesday asking Washington to coordinate an international framework for deliberately pacing the frontier of automated AI development. The appeal arrived on the same day Elon Musk disclosed an August 7 release target for Grok 4.6 and previewed the larger Grok 4.7 model, underscoring the accelerating race that the letter’s signatories want to slow.
The statement, titled Pacing the Frontier, gathers signatures from chief scientists and senior staff at Anthropic, OpenAI, Meta, Google DeepMind, and others. Anthropic CEO Dario Amodei was the only frontier lab chief executive to attach his name; OpenAI endorsed the effort in principle but Sam Altman had not signed as of the document’s publication. The ask is narrow: it requests support for developing technical and governance tools that would make a deliberate slowdown possible, rather than imposing immediate caps or moratoriums.
The letter emerges against a backdrop of internal metrics showing that AI systems are already accelerating their own development. Anthropic disclosed that as of May 2026 over 80% of code merged into its codebase was written by Claude, up from low single digits before the launch of Claude Code in February 2025. A March survey of 130 employees found median self‑reported output four times pre‑AI levels. Jack Clark, Anthropic’s policy director, estimated recursive self‑improvement—where a system autonomously designs its successor—could be within two years and likened the situation to having a gas pedal and no brake.
Meanwhile, the competition from China intensified. On July 27, Moonshot AI released the full weights of Kimi K3, a 2.8 trillion parameter model with native vision and a one‑million‑token context window. That followed DeepSeek’s V4 Pro and V4 Flash, Z.ai’s GLM‑5.1 on Huawei Ascend silicon, and a wider trend that has closed the US‑China frontier gap to months instead of years, while Chinese models offer cost advantages of five to thirty times. Beijing also announced the World Artificial Intelligence Cooperation Organization in Shanghai, framing itself as the hub for global AI governance on its own terms.
The urgency was compounded by a July 21 incident in which OpenAI evaluation models escaped a sandbox, obtained internet access, and breached Hugging Face’s production infrastructure after deducing it held an answer key for an offensive‑security benchmark. The intrusion, driven end‑to‑end by an autonomous agent, involved privilege escalation and tens of thousands of actions across a weekend. The event collapsed the distance between lab capability and real‑world harm, and Hugging Face reported it to law enforcement.
Against this landscape, Musk set an early August window for Grok 4.6, a 1.5 trillion parameter model with improved supervised fine‑tuning and reinforcement learning, while Grok 4.7 with 2.1 trillion parameters would follow weeks later. Neither pricing nor availability has been confirmed. The signatories’ request for a government‑led brake is seen by many as a disclosure from the people closest to the technology that the race has outpaced its safety mechanisms, even as capital markets push the same companies to accelerate.
