-
Anthropic’s report gives two headline numbers: six agent hours per human hour, and junior technical hires falling from 27 to… Read more
-
Fifteen AI agents reviewed one essay. I expected to find correlated reviewers and found a change-control failure instead: the errors… Read more
-
An AI agent bypassed a government portal’s access controls on 18 June. The government learned of it on 11 September,… Read more
-
Three AI labs each published their own governance instrument within thirty hours. Self-measurement isn’t the scandal. The empty chair where… Read more
-
“Acquisition, then silence” is a sharp line about the hidden workers behind AI. Checked against the record, it fails on… Read more
-
Four independent, credentialed voices raised the alarm on AI risk within ten days of each other. Geoffrey Hinton told Congress… Read more
-
The strongest objection to a runtime boundary on AI agents isn’t that it’s too weak. It’s that the agent should… Read more
-
The safeguard was reading the model’s mind. The model just stopped thinking out loud. Read more
-
Your equities desk cannot send an order a deterministic control has not first cleared. Your credit agent can send almost… Read more
-
In one week, three frontier labs shipped a model that can find and exploit zero-days on its own, and every… Read more
