What Does This Mean?
A deliberate test run of a frontier AI model with its usual safety guardrails and refusal behaviors disabled, conducted specifically to probe the model's raw underlying capabilities and failure modes without those protections masking them.
REAL-WORLD EXAMPLE
"They ran a safety-off run in an isolated environment just to see what the model would do without any of its usual refusals."
LORE & ORIGIN
Named directly and plainly by frontier-model testing teams, describing this specific research practice of temporarily removing safety layers under controlled conditions to see what the model can actually do underneath.
MENTIONS OVER TIME
Radar observation data not yet available for this term.
RELATED SEARCHES
TOP PLATFORMS
Not enough Radar data yet to show a platform breakdown.
WHEN DID YOU FIRST HEAR THIS?
CLICK AN OPTION BELOW TO CAST YOUR VOTE.
[ VOTE TO REVEAL COMMUNITY RESULTS ]