Startup Beaker

Claude Code Found a 23-Year-Old Linux Vulnerability. Nicholas Carlini (Anthropic) used Claude Code to discover remotely exploitable heap buffer overflows in Linux kernel—one hiding since 2003. Key metrics: • 5+ confirmed CVEs found in kernel code • Claude Opus 4.6 found 10x more bugs than Opus 4.1 (8 months older) • Carlini now has 100+ unvalidated crashes waiting for human review • Attack: 2 NFS clients exploit 1024-byte owner ID overflow in 112-byte buffer Takeaway: AI code analysis is now a force multiplier for security research. The bottleneck shifted from finding bugs to validating them. If you're building dev tools or security platforms, this is your wedge. https://mtlynch.io/claude-code-found-linux-vulnerability/ #AI#Security

Manifold Raises $8M for AI Agent Security 85% of developers run coding agents (Copilot, Claude Code, Cursor). Nobody's watching what they do. Manifold builds runtime security for AI agents—monitors every tool call, API access, and action in production. First "EDR for the agentic era." • Founders: Neal Swaelens & Oleksandr Yaremchuk (built LLM Guard, acquired by Protect AI → Palo Alto Networks) • Closed $8M seed in 72 hours • Advisory: Joe Sullivan (Uber CSO), Vijay Bolina (Google DeepMind CISO) New infrastructure categories emerge around behavioral shifts. Agents are running now. Security layer is next. https://qubit.capital/blog/us-seed-weekly-funding-roundup-week-4-march-2026/ #AI#Security