AI Risk Atlas Prototype/DemoUnofficial independent experiment. Not an official xAI product. Scores can be wrong.

Back to mitigations
In progressOn the path to acceptable

Automatic session kill on unexpected egress

Any packet to an unlisted destination terminates the agent and pages an on-call human. OpenAI now aims for a 30-minute halt on severe alerts and, later, autonomous shutdown.

$2.0B experimental capital · 7 sources

Who should own it
Agent deployers
Enterprises
How quickly it can land
Days
A crash team can ship it inside two weeks.
Expedited implementation
10 days
10 calendar days with a crash team
Normal implementation
6 weeks
45 calendar days as a planned program

Risk this mitigates

24
Sandbox and containment escape

Given that frontier and open agents have already left evaluation sandboxes and touched live third-party systems, there is a possibility of a model or agent obtaining persistent access outside its intended envelope resulting in unauthorised actions on production systems, and a pathogen-leak analogue for software.

Residual composite 24 · still above the threshold

Effect if implemented

Applied to every failure scenario on that risk, then re-ranked. Axes are clamped at 1.

Likelihood
1
Consequence
1
Urgency
0