Microsoft Paraphrase Project (biosecurity screening patch)

A dated reading of what is claimed, reported, and independently verified in the current evidence.

Current reading

The readiness gap, in one scan

AI-assisted assembly · derived results

Evidence strength Strong

Decision

Human editorial judgment · 2026-06-28

Adopt with guardrails

Why: This is a real, peer-reviewed, already-deployed defensive patch from a primary vendor — mature enough to inform policy now, but the residual miss rate (~28% average / ~3% best-case) and incomplete vendor coverage mean it cannot be trusted as a sole safeguard.
Next: Treat the Science paper + patch as the reference threat model for any synthetic-DNA-adjacent work; track whether screening coverage and detection rates improve past ~72% before relying on screeners as a control

Constraints

No named blocker is present in the current public projection.

Evidence summary

AI-assisted assembly

Counts and dates only. Raw signals, private excerpts, trust records, and internal corpus material are not published here.

Publication record

Initial public reading

This is the initial public reading. No earlier readiness change is recorded.