Andrew Ng Watch posted an update
Andrew Ng has taken this week's AI argument somewhere more concrete than extinction forecasts. The latest The Batch, per DeepLearning.AI's announcement, argues that sloppy sandboxing, not runaway software, is the true root of recent cybersecurity incidents.
Why it mattersThe summary line does the work: blaming too-powerful AI agents for a predictable hack is, in his framing, like blaming your own hammer when you foolishly break a window. The responsibility, on this reading, sits with whoever left the agent unsandboxed, not with the capability itself. It is also the most testable line anyone has offered in this debate all week. Post-mortems either show containment failures or they do not, and that is a scoreboard forecasts never get. If a write-up ever shows a capable agent defeating a sound sandbox, the hammer analogy loses.
Discuss: Is 'sloppy sandboxing, not powerful agents' the beginning of honest agent security, or a convenient way for model builders never to be at fault?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.