Final on Nostr: This won't entirely replace human reviews yet. The high point of an AI agent's ...

This won't entirely replace human reviews yet. The high point of an AI agent's capabilities is only slightly over the capability of it's user. Actual research orgs have vulnerability disclosures exponentially increase with AI while an average dev will mostly just close out simple security deficiencies. Creating a PoC for exploiting a vuln is also resourceful.

Serious vulnerabilities held under embargo on Android have increased massively. You can see our changelog on the security preview patches per release and see how the list gets bigger even when some vulnerabilities get removed from the list as they become part of AOSP.