OpenAI ships 1.5B privacy filter

- OpenAI rolled a 1.5‑billion‑parameter privacy filter designed to redact or block personally identifiable information before model outputs leave a device or pipeline. - The 1.5B privacy filter was presented as a local redaction layer that can run closer to users to limit PII exposure in downstream logs. - The release sparked debate because independent sampling shows fewer than 20% of deployed AI agents currently publish safety policies, raising transparency worries (x.com 1) (x.com 2).