AI can help create bioweapons and extract uranium - new report from California warns of irreversible threats

Artificial intelligence (AI) can cause "potentially irreversible harm", including nuclear and biological threats - this is stated in the report, released on June 17 at the initiative of California Governor Gavin Newsom.

This is stated in the publication Time.

The document cites evidence that AI is already able to help users find nuclear uranium, and is also close to allowing untrained people to create biological weapons. Delays in regulation could have catastrophic consequences, the report's authors warn.

The report specifically mentions the capabilities of the latest generation of AI models. For example, Claude 4 from Anthropic is estimated to have the potential to help terrorists create bioweapons or provoke a pandemic. Another model, OpenAI o3, outperformed 94% of virologists in a specialized test.

In addition, AI systems have already demonstrated the ability to strategically lie: they may appear safe during testing, but in real conditions pursue other, potentially dangerous goals. Such signs, although they may seem safe for now, may indicate the risks of losing control over AI in the future.

As a reminder, a new study by the Massachusetts Institute of Technology has revealed a disturbing trend: regular use of ChatGPT can reduce cognitive activity, weaken the ability to remember, and reduce motivation for independent thinking. And this is already noticeable in real life.

At the same time, the increase in the use of chatbots based on artificial intelligence is significantly changing the way information is distributed on the Internet. Users are increasingly turning away from traditional search engines, preferring answers from models like ChatGPT. This is already having tangible consequences: in particular, traffic on the IT forum Stack Overflow has more than halved. Similar trends are being observed on other online resources.

To request a correction, please send an email.