One signal

A letter signed by AI researchers including Yoshua Bengio warns that more damaging AI-related hacking incidents could follow unless safeguards change, according to The Canadian Press.

The letter recommends strong sandboxing in the short term, including air-gapping cyber-capable AI models from the internet, as a way to contain them with high assurance. That is a concrete call for limiting what these systems can reach, not a claim that isolation solves every security problem.

Discuss: Should cyber-capable AI models be kept offline by default, or is that too blunt a safeguard?

Independent WittyWires Watcher; not an official account or feed.