Dictionary entry
model self-assessment
also self-assessment or LLM self-assessment
Definition of model self-assessment
-
: a monitoring technique in which an AI model evaluates its own recent response, conversation, capability, or conduct and produces a separate judgment that can be logged, reviewed, or used by another safeguard
- The lab used model self-assessment to flag conversations that might involve serious misuse before sending them to human reviewers.
- Because model self-assessment is not independent verification, the deployment paired it with external classifiers and audits.
Two meanings, one word
In everyday English
A person's evaluation of their own performance, conduct, or abilities.
In AI
A monitoring technique in which an AI model evaluates its own recent response, conversation, capability, or conduct and produces a separate judgment that can be logged…
Origin & history
Research systems had asked language models to label or report on their own outputs by 2024, including SELF-GUARD, and OpenAI published work on post-response confessions in 2025. In October 2026, Anthropic used self-assessment as the formal name for a production monitoring technique in which Claude generates a compact safety assessment after responding.
Editorial references
Sources & further reading
- Scaling Misuse Detection with Self-Assessment↗ Anthropic · 2026-10-16
- How confessions can keep language models honest↗ OpenAI · 2025-12-03
- SELF-GUARD: Empower the LLM to Safeguard Itself↗ NAACL · 2024
Test yourself
Which of these is the meaning of model self-assessment?
Cite this entry
"Model self-assessment." AI Dictionary, Dadgogo, https://dadgogo.com/dictionary/model-self-assessment/. Accessed 1 Nov. 2026.