Dario Amodei, CEO of Anthropic, has called on other AI companies to conduct mandatory safety testing of neural networks before their public release, according to Bloomberg.
“I think we should make testing mandatory, but it’s also important to be very careful about how we do it,” he said.
The head of the AI startup noted that there are a number of voluntarily adopted safety rules agreed upon by major firms. However, clearer standards are necessary.
“There is nothing to verify or ensure compliance with the rules from the company’s side. They just said they would do it,” he emphasized.
Amodei added that startups have adhered to certain standards under public and employee pressure, but ultimately this will prove insufficient. At the same time, testing requirements should remain “flexible” to accommodate the pace of technological change.
In November, Anthropic opened its AI products to U.S. intelligence and defense agencies. Earlier, Meta took a similar step.
Earlier in the same month, Amodei stated that artificial intelligence would reach human-level capabilities by 2026.