Anthropic Chief Calls for Slower AI Progress to Keep Safety on Track

Swan.my.id - Jakarta, Anthropic CEO Dario Amodei has called for slower AI development, warning that safety efforts could fall behind as artificial intelligence capabilities advance rapidly.

Amodei said the industry should avoid moving forward without allowing enough time to address potential risks. He does not support stopping AI development entirely, but wants companies to manage the pace more carefully.

He made the appeal in a lengthy essay titled “We Must Pace the Frontier,” published recently on his personal blog. Amodei argued that safety and alignment efforts need time to catch up with increasingly capable AI models.

Amodei Warns Safety Work Could Fall Behind

According to Amodei, AI capabilities are developing at a pace that creates new challenges for researchers and companies. If the industry refuses to “hit the brakes,” human efforts to keep AI safe could struggle to keep up.

“Over the last few months, I’ve become increasingly convinced that to really address AI risk, we need greater caution... while also pacing the rate of capability improvement so that risk mitigation efforts have time to catch up,” Amodei wrote.

His comments came only a few days after OpenAI Chief Scientist Jakub Pachocki raised similar concerns. The warning also received support from leaders of companies that compete directly with Anthropic, including OpenAI CEO Sam Altman and xAI founder and CEO Elon Musk.

Slower AI Development Would Not Mean a Full Halt

Amodei’s proposal does not call for a complete suspension of AI research. Instead, he wants companies to develop the technology at a speed that allows safety systems to advance alongside model capabilities.

One priority is alignment, or ensuring that AI systems remain consistent with human interests. He also stressed that oversight mechanisms should become more capable as AI models grow more advanced.

This approach would give companies and independent experts more time to assess risks before new systems become widely deployed. It would also create room to identify problems during training and evaluation.

Three Steps Proposed by Anthropic’s CEO

Amodei outlined three measures for managing AI risks. First, companies should provide enough time to secure and align their models while giving independent evaluators access to the systems.

Second, leading AI companies should coordinate to create shared safety standards. Third, he called for international coordination to manage risks that could extend beyond the borders of any one country.

• Allow sufficient time for model safety and alignment work.
• Give independent evaluators access to assess AI systems.
• Create shared safety standards among leading AI companies.
• Strengthen international coordination on AI risks.

Anthropic Plans Permanent Third-Party Access

Anthropic said it would carry out the first step independently by giving third-party evaluators permanent, employee-level access to its systems.

The independent parties would be able to examine compliance with safety standards, report incidents, and evaluate model alignment during training. The plan reflects Amodei’s broader argument that external oversight should develop alongside AI capabilities.

Amodei’s position adds to growing calls for greater caution within the AI industry. While companies continue competing to build more capable systems, the debate now also focuses on whether safety research and oversight can move at the same speed.