Skip to content
Vol. I · No. 251
Mon · 8 Jun
A Daily Lexicon of Trustworthy Data
The Lexicon

006·342

guardrails

/ˈɡɑːrdreɪlz/ - n.

1 [colloq.] Rules that enforce a policy, written for the policy the org has not yet enforced anywhere else.Keep. Punchy.This is the problem.

Working definition

2. Programmatic constraints that filter or block model inputs and outputs against defined safety and policy rules.

Filed
See also
  • ai governanceTerm struck pending the decision rights, review cadence, and accountability for training data and outcomes it presumes someone already assigned.
  • ai policyA document that prohibits the thing three teams shipped last quarter, dated next quarter.
  • hallucinationThe model answering a question the organization had also never answered, only faster.
  • prompt injectionDiscovering the assistant trusts the document more than the policy that approved it.