The GitHub repository "exploitarium" contains a collection of public exploit proof-of-concepts (PoCs) and vulnerability research writeups, with a focus on serious vulnerabilities. AI was utilized to automate the fuzzing workflow using GPT-5.5-3-Codex-Spark, allowing for efficient testing with minimal manual intervention.

Anthropic's Mythos and Project Glasswing suggest that advanced AI vulnerability research should be restricted, but replication studies show that similar capabilities are already present in public models. Frontier models are increasingly effective at identifying serious vulnerabilities in real software.

There has been a significant increase in daily reports of duplicate vulnerabilities, with multiple individuals identifying the same bug using different tools. This trend marks a shift from previous inconsistencies in vulnerability reporting.

AI coding agents are expected to significantly change the landscape of exploit development and vulnerability research in the coming months. Improvements in frontier models will lead to rapid advancements rather than gradual changes, resulting in a surge of high-impact security vulnerabilities.