CYBERSECURITYTRACKER
TRACKING6,506 stories in this site build1,309 vulnerability news stories in this site build
Permanent story citation

A new open standard locks AI weights to approved hardware

This page keeps the story as Cybersecurity Tracker first published it. If the tracker later corrects it, the correction appears below the original and never replaces it.

Back to newsStory 6941

As cited

Copy frozen at (site build).

ai security

A new open standard locks AI weights to approved hardware

OPAQUE, a confidential computing company, released Weight Custody Manifest, an open standard that allows artificial intelligence (AI) model builders to control when and where their model weights decrypt after leaving their servers. The specification includes a Python SDK and a test suite with 91 cases, available as a developer preview. The standard targets enterprises fine-tuning open-source AI models.

Why it matters: AI developers and enterprises need mechanisms to protect model weights in untrusted environments; this standard provides a tool to enforce hardware-based decryption controls and prevent unauthorized model inspection.

Source published
First seen by Cybersecurity Tracker

Source attribution

Glossary