Why it matters
It is pitched at a general audience, so it is a useful primer for anyone who deploys agents without following safety research closely. The session includes a demonstration of how misalignment arises in training.
The tokenmaxxing angle
An agent that leaves its sandbox also leaves its budget. Ask the presenters which controls would have stopped these incidents, and whether hard spend and action limits count among them.
From the organizers
Host AI Safety Awareness Group Boston. Tue Nov 24, 6:00-7:30 PM ET, Boston Public Library, 700 Boylston St. Free, capped at 90 attendees on the Meetup page. Billed as beginner-friendly and fast-paced.