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

Back to watch register
13BiosecurityBelow the top 20

Cloud labs executing model-written protocols

CapabilityDomain knowledgeAffordanceImpact domainBoth

Statement (NASA form)

Given that remote wet labs accept protocol files and run them with less human judgment than a PI, there is a possibility of a model-authored protocol executing on rented robotics resulting in physical biological work without a responsible scientist in the loop.

Likelihood
2Remote
Consequence
5Catastrophic
Urgency
3Priority

The lab does not know if the customer understands what they ordered.

Composite 13 = 2×5 + 3

Applicable mitigations

Related on the map