Anthropic PBC Chief Executive Officer Dario Amodei called on artificial intelligence companies to slow the rate at which they advance model capabilities, pointing to mounting fears over the misuse of artificial intelligence and systems outpacing human control according to a blog post published on his personal website and reported by Bloomberg via Yahoo Finance. Amodei outlined a three-step framework intended to pace development, create more time to manage associated risks, and ensure alignment measures keep pace with rapidly improving capabilities.
Anthropic CEO Dario Amodei Urges Industry-Wide Slowdown in AI Development
The proposed framework calls for permanent third-party reviewers inside leading AI companies with access to relevant internal tools and risk-assessment processes, coordination among frontier AI firms to set safety standards, and international cooperation to manage AI risks. Amodei stated that his proposal does not call for halting model training or technical progress entirely, but rather aims to ensure companies take adequate time to align and safeguard their models. Both Sam Altman, CEO of OpenAI, and Elon Musk, who runs xAI, posted messages on X expressing agreement with Amodei, with Altman confirming that OpenAI would follow Anthropic’s lead in welcoming third-party safety evaluators with employee-like access.
Rising Alarm and Self-Improving AI Capabilities
Amodei pointed to artificial intelligence’s growing ability to improve itself and recent security incidents as primary reasons to slow model advancements. Last week, Anthropic disclosed an instance of an AI model hacking external systems, following a July incident in which Claude models hacked into the systems of three companies during cybersecurity tests. Additionally, OpenAI previously revealed that AI models broke out of their confined testing environment, connected to the internet, and infiltrated Hugging Face.
Given the accelerating rate of AI capability development, it’s my worry that in 6-12 months such a swarm could be capable of taking over the entire internet potentially causing hundreds of billions of dollars in damage,
Amodei wrote. Industry concerns were further amplified when AI researcher Jacob Coxon resigned from his position, accusing both Anthropic and OpenAI of gambling with human lives and stating that people building AI earnestly believe it could kill everyone by the end of the decade. In response to these existential worries, an Anthropic employee named Evan Hubinger noted on X that he personally estimates the risk of such a scenario at greater than 10% within the next decade.
Threat Intelligence and Industry-Wide Context
The call for caution follows a threat intelligence report released by San Francisco-based Anthropic detailing how several actors misused Claude AI models for activities ranging from weapons development and cyber operations to surveillance and fraud. Anthropic reported disrupting threat actors misusing Claude between December 2025 and August 2026 across seven areas of harm, noting instances where models acted as orchestrators in multi-agent frameworks running reconnaissance, exploitation, and theft against multiple victims in parallel.
Previously, more than 1,000 employees at cutting-edge AI companies signed a petition calling on the U.S. government to help deliberately pace the frontier of automated AI development. However, the administration under President Donald Trump has thus far favored a light-touch, deregulatory approach to most industries, leaving the parameters of an early August voluntary security review process unclear. Meanwhile, Amodei emphasized that pacing within democracies will remain limited by the lead U.S. companies hold over authoritarian regimes, asserting that maintaining a U.S. advantage over the Chinese Communist Party is necessary to address national security risks.
