Call to Slow Down Trust and Development Processes in the Artificial Intelligence Sector

Serdar HocamAuthor & Editor

While leading U.S. artificial intelligence laboratories agree that development must be slowed down to increase safety, experts emphasize that legal standards are essential.

◉ 0 views
Building Trust in AI Is Critical to the Frontier’s Future | Council on Foreign Relations

Executives of leading U.S. artificial intelligence companies have reached a rare public consensus that development should be slowed down in order to improve safety and oversight mechanisms. Experts state that going beyond voluntary commitments and establishing independent audits and legal standards is of vital importance.

Slowing Down Development in Artificial Intelligence Laboratories

Leaders of prominent artificial intelligence laboratories such as Anthropic, OpenAI, SpaceX, and Google DeepMind have agreed that the sector should slow down its development to ensure adequate oversight and safety controls.

Independent Evaluations and Transparency Steps

Anthropic and OpenAI have committed to providing staff-level access to independent third-party evaluators to examine safety risks related to model development and behavior. These companies announced that they will allow the unedited publication of the evaluation results.

Legal Standards and Self-Regulatory Body Proposal

The establishment of an artificial intelligence self-regulatory body authorized by Congress will legalize the role and enforcement authority of independent evaluators, increasing cooperation in combating safety challenges.

The Role of Critical Sectors and Buyers

Companies operating in regulated sectors such as finance, health, and life sciences should demand minimum levels of vital safety guarantees in their artificial intelligence integration processes.

Open-Weight Models and Cybersecurity Collaboration

The Open Secure AI Alliance, consisting of over a hundred companies led by Nvidia, is conducting joint efforts to identify and mitigate security vulnerabilities in widely used open-weight models.