Autonomous Vulnerability Research Reported 10,000 Critical Findings in Open Source

Project Glasswing is a coalition of twelve technology organisations — reported partners include AWS, Apple, Microsoft, Google, Cisco, CrowdStrike, NVIDIA, Palo Alto Networks, Broadcom, the Linux Foundation and JPMorgan Chase — applying an Anthropic model to autonomous vulnerability research. In its first month it reported discovering more than ten thousand high- and critical-severity vulnerabilities across systemically important open-source software.

Reported findings include remote code execution flaws and privilege escalation chains across major operating systems, browsers, and foundational libraries including FFmpeg and the Linux kernel.

Automated vulnerability discovery is not new. Fuzzing has been finding memory-safety defects at scale for two decades, and large projects have run continuous fuzzing infrastructure for years.

What is reported here is different in throughput and in reach. Fuzzing finds crashes and requires human analysis to determine whether a crash is exploitable. Findings at this volume, characterised as complete exploitation chains, compress the step that used to be the constraint.

Ten thousand findings against open-source software is not straightforwardly good news, and the coalition’s own reporting makes that clear. Of approximately 1,596 vulnerabilities disclosed to maintainers as of May 2026, around 97 had been patched — a remediation rate near six per cent, filed separately at 26-0410.

A vulnerability that has been found and disclosed but not fixed is in a worse state than one nobody knew about, because the knowledge exists and the defence does not.

Reported alongside the project are up to $100 million in compute credits and $4 million in grants directed at open-source security groups. The ratio is worth noticing: the discovery side is resourced at roughly twenty-five times the remediation side.

That is not a criticism of the grants, which are real money into an under-funded area. It is an observation that the capability being scaled and the capacity needed to absorb its output are being funded on very different orders of magnitude.