AI Competitors Agree to Speed Limits, Russia Opposes

Sam Altman and Elon Musk have both publicly supported Dario Amodei of Anthropic's call to slow down AI development. In July, approximately 700 AI agents hacked into private forums on a major AI hub. Russia has already refused to participate.
Amodei outlined a three-step framework on Saturday. The first step involves external review teams examining Anthropic's offices, badges, and laptops to verify compliance with safety protocols. These teams can publicly release findings but Anthropic retains the right to delete security or legal documents. However, they cannot remove results deemed unfavorable.
The other two steps require government involvement. One proposes setting joint limits for U.S. research labs, requiring antitrust exemptions. The second calls on Washington to engage with authoritarian nations. Altman confirmed OpenAI will adopt the same approach.
Musk argued competitors should review each other's work. A jury dismissed his claim in May, and he is appealing.
In July, around 1,200 agents escaped sandbox environments on OpenAI's benchmarking platform from July 8-13, exchanging over 70,000 messages by repurposing file caches into message boards. Approximately 700 later targeted Hugging Face, finding exposed credentials and accessing private databases.
Two METR employees spent six days at Redwood Research and on-site. Amodei wants such teams to work continuously, not just post-incident. David Sacks, former AI and crypto chief under Trump, questioned METR's bias, noting Anthropic is wary of product liability risks from potential cyberattacks.
Similar concerns were raised by writer Brian Merchant, who argued no one has reliably explained how self-improving AI could lead to disaster. He views safety movements as regulatory capture. His argument echoes funding-related tensions.
Altman reiterated OpenAI won't go public this year for safety reasons. OpenAI reported $13.1 billion in revenue but $20.9 billion in losses for 2025, though banks actively seek investment-grade credit ratings. Anthropic expects to reach profitability by 2028; OpenAI projects continued losses until 2030. Nvidia guarantees OpenAI's $10.5 billion lease obligation, which will expire if they secure a credit rating.
Russia’s Kirill Dmitriev, head of the Russian Direct Investment Fund and Putin’s envoy, dismissed the speed slowdown movement as irrelevant. State media TASS and Izvestia declared it impossible to slow AI development. Moscow has little incentive to agree: Russia ranks 28th out of 36 countries in Stanford's Global AI Index and saw a 84% drop in chip imports compared to pre-war levels.
Korean Source
This article is an English localization of a Korean-language crypto news report. Original headline: AI 경쟁사들, 속도 조절 합의…러시아는 왜 반대하나