Tsinghua's BEMULA Benchmark Reveals AI Agents Almost Never Truly Forget

Tsinghua researchers introduced BEMULA (BEMU), the first benchmark evaluating memory unlearning in LLM agents, finding that when users ask agents to delete personal information, virtually no system succeeds, with end-to-end pass rates below 1.5%.

2026-11-04 ~ 2026-11-04 · 2 related posts

• Tsinghua's BEMULA benchmark: no LLM agent memory system passes over 1.5% of forget requests — rohanpaul_ai · 2026-11-04

1 near-duplicate retellings: rohanpaul_ai