World ↗Civilization build(t) · 09/10/2026
The current build
AI-assisted vulnerability discovery has reached a scale where human verification and remediation are visibly lagging candidate generation in a large open-source security program. Anthropic reports 29,439 candidate findings and 6,123 externally triaged findings as of October 2, with 5,674 confirmed valid; maintainers have already patched a subset and independent project records confirm fixes originating from the program. The shift is real but bounded: it does not show that all AI-generated findings are correct, that remediation is automated, or that human verification has become the dominant bottleneck across every domain.
Claims — click “why?”
Anthropic's open-source vulnerability program provides concrete evidence that model-generated vulnerability discovery can outpace available human triage capacity, with external review confirming a large share of reviewed findings while remediation remains substantially downstream.
supported inference 96% Why?
Why this claim?
The throughput gap may be specific to Anthropic's program design and disclosure process rather than a general human-AI bottleneck.