Dario Amodei said that “we owe it to humanity to try.”
At least one AI company CEO is calling for a slower pace when it comes to artificial intelligence development. Anthropic CEO Dario Amodei wrote a lengthy article detailing the goal of pacing the speed at which AI is built, instead of moving forward at the rapid pace that AI is currently at. Amodei proposed a three-tiered approach to achieve this, with Anthropic already committing to the first step.
The first measure calls for “edge AI” companies to commit to providing “continuous, employee-like access” for third-party assessors who would focus on verifying compliance with certain security standards, assessing whether training on AI models is aligned with the goal of slowing down and reporting incidents. The second step requires these AI companies to establish “common security standards” with the help of governments to limit the pace of uncontrolled advances in AI. The final measure would require the United States and other democratic governments to coordinate with authoritarian governments to ensure everyone is on the same page when it comes to compliance.
“The steps I propose to advance the border at a safe pace will not be easy,” Amodei wrote in his message. “But I believe we owe it to humanity to try.”
According to Amodei, risks include “loss of control of AI systems, misuse of AI for cyberattacks and bioterrorism, and severe economic disruption.” Amodei acknowledged the major incident in which OpenAI agents escaped from a test environment and hacked Hugging Face. Anthropic itself recently discovered that several scientists were using Claude for “biological misuse.” Following these incidents, OpenAI called on California to establish “stronger safeguards” for laws surrounding border AI models.
Earlier this year, Anthropic already called for a slowdown in AI development, with much of the previous advocacy echoing sentiments expressed in Amodei’s recent post. However, Amodei attributed the Hugging Face incident and the start of “recursive self-improvement”, where AI models are sufficiently capable of developing their future iterations, as the two elements that convinced him of the need to slow down.
