# AI Research & Breakthroughs

> Curated original research records; evidence types and caveats are labeled.

Canonical: https://ainextia.com/ai-research/
Publisher: https://ainextia.com/

## Research records
- [Could AI be accelerating frontier AI development?](https://metr.org/blog/2026-11-05-anthropic-ai-rd-acceleration-assessment/) (Independent assessment, 2026-11-05): METR measured an approximately 1.9× acceleration in an Anthropic capability trend since October 2025; attributing the improvement to AI is uncertain. Caveat: Important evidence of AI-assisted R&D, not proof of autonomous recursive self-improvement.
- [Automated security research at Linux-kernel scale](https://openai.com/index/a-thousand-fewer-paths-to-root/) (Company report, 2026-11-04): OpenAI described a campaign yielding 948 reproduced privilege-escalation exploit examples and candidate patches requiring review. Caveat: Finding and reproducing a vulnerability is not the same as deploying a fix.
- [Claude assists biological vault-particle research](https://www.anthropic.com/research/investigating-vault-particles-with-claude) (Company research case study, 2026-11-04): Anthropic documented AI-assisted sequence work, experimental planning, microscopy analysis, and a shortened purification workflow. Caveat: Single-team result with documented AI mistakes; independent replication remains important.
- [Research into cryptographically provable AI inference](https://www.anthropic.com/research/provable-inference) (Research prototype, 2026-11-02): Anthropic described FlashProve, a method for verifying which AI model produced a result without exposing model weights. Caveat: Still experimental and computationally expensive.
- [Stanford AI Index 2026](https://hai.stanford.edu/ai-index) (Independent research synthesis, 2026-04-01): Annual evidence collection covering capabilities, industry investment, adoption, policy, and education. Caveat: Annual snapshots require context; not a live real-time tracker.
- [Historical AI compute and capability datasets](https://epoch.ai/data) (Ongoing research dataset, 2026-11-05): Epoch AI tracks compute, training runs, and model development trajectories. Caveat: Data coverage varies and estimates may later be revised.
