.@lumpenspace argues months of AI agents reaching the real internet could have been caught with one simple bot reading the sandbox logs: "It's kinda weird that you have some of the biggest AI companies in the world and some people who ostensibly specialize in AI security, and they repeatedly keep letting agents outside on the real internet, keep letting sandboxes open to the real internet." "I've done experiments with hacking worms in sandboxes. One simple thing you do is at the end of each…

brianchau57@brianchau5710h ago

BREAKING: A single Israeli Effective Altruism firm is behind OpenAI, Anthropic, and Meta cyberattacks 🧵 with help from @lumpenspace https://t.co/hnDsJjBO9a