Import AI 470: No rights for machines; automating environment generation with SPADE; and building better GPU kernels with Hawkeye

This Import AI issue covers a METR analysis finding AI-driven acceleration in cybersecurity, modest gains in math, and no measurable acceleration in AI research optimization; SPADE, a self-play framework that co-evolves executable training environments and agents, improving Qwen3 backbones on game and tool-use benchmarks; and Hawkeye, an open-source framework of unit-test taxonomies that helps coding agents generate hardware-aware GPU kernels, reaching an 18.9x geomean speedup over expert Triton kernels on emerging attention variants. It also includes a researcher's essay on fearing AI's implications for human meaning, an argument against granting AI systems rights, and DeepMind's use of AlphaEvolve to improve the matrix multiplication exponent.