Researcher on Safeguards team at Anthropic
Experimented with agent teams where multiple Claude instances work in parallel on shared codebases, tasking 16 agents with writing a Rust-based C compiler from scratch.
How media typically covers Nicholas Carlini
Directly quoted in these articles
OpenAI and Anthropic conducted rare joint safety testing by granting each other API access to models with reduced safeguards, demonstrating how competing AI labs can collaborate on safety despite intense competitive pressure.
“Anthropic safety researcher expressing interest in continuing OpenAI access for future safety research collaboration.”
Referenced in coverage
16 parallel Claude instances working autonomously on a shared codebase produced a 100,000-line Rust-based C compiler capable of compiling the Linux kernel across x86, ARM, and RISC-V architectures.
“Experimented with agent teams where multiple Claude instances work in parallel on shared codebases, tasking 16 agents with writing a Rust-based C compiler from scratch.”
“Co-authored research on AI models' ability to exploit smart contracts and cyber capabilities.”