
Catching spam that hides behind misspelled words
Discover the storyResult & evidence
RETVec helps email filters recognise unwanted messages even when senders distort words or insert unusual characters. This makes a common protective service more resilient while keeping legitimate correspondence accessible to users.
The research evaluation detected 38% of spam missed by the baseline model, at the stated false-positive threshold.
Google Research; Gmail
Global 2023
Provider-run evaluation on internal data; the 38% refers to previously missed spam, not all incoming mail.
- RETVec: Resilient and Efficient Text Vectorizer Bursztein and colleagues, NeurIPS 2023 · 2023
- Improving Text Classification Resilience and Efficiency with RETVec Google Security Blog · 2023-11-29
- RETVec: implementation and archive notice Google Research · undated


