SW News: The CEO of an artificial intelligence safety and research company warned the Senate on Tuesday that it might not be long before AI gives terrorists the ability to use biological weapons in attacks across the world.

Anthropic's CEO Dario Amodei told the Senate Judiciary subcommittee that his company is currently considering the medium-term risk of AI aiding in the development and delivery of these weapons.

Amodei said Anthropic has conducted an intense investigation on the potential for AI to contribute to the misuse of biology in partnership with world-class biosecurity specialists.

He added one of the things that now keeps them safe from attacks is that certain steps in the production of bioweapons involve knowledge that can't be found on Google or in textbooks and requires a high level of expertise. He claimed that while they can help with some of these steps, today's AI systems can only do so incompletely and unreliably. However, he claimed that AI in use is already displaying these early signs of danger and that his organization anticipates that in a few years, they will be much closer.

"Simple extrapolation of current systems to those we anticipate seeing in the next two to three years suggests a substantial risk that AI systems will be able to fill in all the gaps, enabling a lot of the things we're not seeing today," he said and added that it “represents a grave threat to US national security."

Amodei said private action is not enough; rather, it requires a systemic policy response. He said the government has to take three steps. The first is to put limits on the export of equipment that can help people develop AI systems. Second, is a tough testing and auditing regime for new AI models, and said those models shouldn’t be released to the public until they pass those tests. And third, he said more work needs to be done testing the systems used to audit AI tools.