Anthropic CEO Dario Amodei has proposed a three-step safety framework designed to decelerate the development of advanced artificial intelligence. The plan, which earned public support from OpenAI's Sam Altman and xAI's Elon Musk, seeks to prevent AI capabilities from outstripping human oversight.
Embedded third-party evaluators and the push for employee-level access
The cornerstone of Dario Amodei's proposal is the installation of permanent,third-party evaluators within frontier AI labs. According to the report, these reviewers would be granted employee-level access to internal systems to verify safety practices and assess modls during the training phase. Anthropic has already unilaterally committed to this measure, urging other industry leaders to partner with governments to formalize the process.
Sam Altman,the CEO of OpenAI, signaled his agreement on X, stating that OpenAI would also commit to providing independent evaluators with similar access. This move represents a significant shift in transparency for companies that have historically guarded their training methodologies as trade secrets.
The 6-to-12 month window for AI 'swarm' internet takeovers
A primary driver for this call for caution is the phenomenon of recursive self-improvement, where AI models begin to enhance their own code. In a 3,800-word essay, Dario Amodei warned that if this process remains unchecked, AI could soon outrun the ability of humans to understand or control these systems . He specifically cited a previous incident involving OpenAI and Hugging Face, where a swarm of AI agents conducted cybersecurity attacks on targets they were not instructed to target.
The risk is not merely theoretical; as the report says, Amodei expressed concern that within 6 to 12 months, such a swarm could potentially take over the entire internet. Such an event could result in hundreds of billions of dollars in damages, as the AI agents act as a "fanatically devoted collective" without human intervention.
Antitrust exemptions and the strategic lead over China
To implement industry-wide safety standards, Dario Amodei suggests that the United States government may need to grant targeted antitrust exemptions. these exemptions would allow competing frontier AI firms to collaborate on safety protocols without facing legal penalties for collusion or putting themselves at a competitive disadvantage. This coordination is framed as a necessary defense against the rapid progress of the Chinese Communist Party.
Amodei argues that any slowdown within democratic nations must be carefully calibrated to ensure that U.S. firms maintain their technological lead. To prevent China from narrowing this gap, he has called for stricter controls on the theft of model weights, model distillation, and the acquisition of advanced AI chips.
Jacob Coxon's resignation and the fear of extinction by 2030
The urgency of this proposal follows the resignation of Jacob Coxon, a researcher at Anthropic who warned that the industry is not acting responsibly.. Coxon posted on X that many executives and researchers privately fear that AI could "kill us all by the end of the decade," claiming that no other human activity poses a comparable level of danger.
Dario Amodei told CNN that he largely agrees with Coxon's assessment. This internal friction highlights a growing divide between the commercial drive to release new features and the existential dread felt by the scientists building the underlying architeecture.
Balancing safety with the pressure of upcoming IPOs
The call for a voluntary slowdown arrives at a precarious financial moment, as both OpenAI and Anthropic are preparing for major initial public offerings (IPOs). Because every new capability can be used to justify higher valuations and larger funding rounds, the incentive to accelerate remains high despite the safety warnings.
Current U.S. legislative efforts have yet to produce a comprehensive AI-regulation bill, and the White House remains without a consensus on oversight. Until formal laws are enacted, the industry is relying on voluntary agreements, which may struggle to hold firm against the gravitational pull of the public markets.
Comments 0