Open Source Projects and AI Security
summary
gRPC teams adopt the open-source Mantis Skills harness to reduce an early >90% false positive rate from point‑and‑shoot models, adding an ADK reference install, SQLite outputs and an 80K eval to automate six months of scans that exposed protocol mismatches, multi-implementation bugs and clusters caused by overly complex code. Proactive per‑CL scanning and sandbox escape checks accompany fixes after a 12x surge in reports; Code Mender provides a supported remediation path.
AI-written summary · about
chapters (12)
- 0:00Introduction to AI security
- 1:04AI vulnerability discovery reality
- 2:09Mantis open-source harness
- 3:33Design choices in Mantis
- 6:01Vulnerability reporting surge
- 7:07Collective knowledge in security
- 8:30Sandbox escape testing
- 9:33gRPC scanning results
- 13:06Bug categories and patterns
- 14:50Future of vulnerability scanning
- 16:40Resources and next steps
- 17:04Q&A session
related talks
26mOpen Source Tools To Empower Ethical and Robust AI Systems
43mApplication Security is a Community Effort
21mNavigating the Intersection: AI’s Role in Shaping the Secure Open Source Software Ecos...
48mAI Impact on Cybersecurity and Privacy
01h 01m"Looks Good to Me": A Practical Guide to Handling AI-Generated Code
16mPerformance improvements: exposing hidden security fixes in Open Source
44mSecurity Research with Open Source Software: Vol. Omega
34mBalancing Security and Innovation in Multi-Agent Systems: An Open Source Approach