PrivDrift Benchmark

What is PrivDrift?

The benchmark evaluates whether user-disclosed secrets remain recoverable in active LLM conversations after topic drift and persuasion-based probing.

Released

2026-09-24

Evaluates

General AI, Language & Knowledge, cs.AI

Openness

unknown

Importer

Claire Radar

Review status

ai-name-audit-deferred

Reported scores

0

Source provenance

• Original evidence https://arxiv.org/abs/2609.30094

PrivDrift paper

• Paper https://arxiv.org/abs/2609.30094

Which sources cite PrivDrift?

• PrivDrift

Related General AI, Language & Knowledge, cs.AI benchmarks

• $\tau^\tau$-Bench: An Environment for End-To-End, Realistic Agent Construction
• 2d-fet-bench
• 3DCodeBench
• 5-Dialects-BN
• A Large-Scale Dataset and Benchmark
• A Real-World CAD Segmentation Benchmark
• A3S-Bench
• AARRI-Bench