"1,000 days left" Anthropic founder

2026-05-11 · Source: Wes Roth · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Economic Analysis & Policy, Public Policy & Governance · Depth: Novice, extended

Summary

Jack Clark, co-founder of Anthropic and former head of policy at OpenAI, predicts a 60%+ likelihood of fully automated AI research and development (R&D) by the end of 2028. This means AI systems will autonomously build their own successors, leading to recursive self-improvement (RSI) or an "intelligence explosion." Evidence supporting this includes the rapid advancement in AI coding capabilities, exemplified by benchmarks like Sweet Bench and the performance of models such as Claude Mythos Preview, which achieved 93.9% on real-world software engineering problems. AI is also excelling at long-horizon tasks, core science skills like experiment replication (e.g., Opus 4.5 solving Core Bench at 95.5% by December 2025), and kernel design optimization. Google DeepMind has even hired a Director of AGI Economics, indicating serious preparation for AI's profound economic impact, including the potential for a "capital-heavy, human-light" economy with AI-run corporations.

Key takeaway

For AI Scientists and Research Scientists weighing future research directions, prioritize AI alignment and control mechanisms. The rapid progression towards automated AI R&D by 2028 means current alignment techniques may become obsolete as AI systems surpass human supervision. Focus on developing robust, verifiable alignment strategies that can withstand recursive self-improvement, and consider the economic implications of a capital-heavy, human-light economy driven by autonomous AI corporations.

Key insights

Automated AI R&D, where AI autonomously improves itself, is likely by 2028, profoundly reshaping society and the economy.

Principles

• AI capabilities are advancing faster than alignment solutions.
• Emergent properties drive AI progress beyond explicit design.
• Understanding cannot be outsourced or automated.

Method

AI systems are demonstrating self-improvement through optimizing hardware designs (TPUs), enhancing training processes (Gemini), and excelling in coding, scientific experimentation, and kernel optimization.

In practice

• Monitor AI coding benchmarks like Sweet Bench for capability shifts.
• Investigate AI's emergent properties in complex tasks.
• Prepare for economic shifts towards AI-driven corporations.

Topics

• Automated AI R&D
• Recursive Self-Improvement
• AI Alignment
• Economic Disruption
• Claude Mythos

Best for: AI Scientist, Research Scientist, Director of AI/ML, Policy Maker, General Interest

Related on AIssential

• Import AI 455: AI systems are about to start building themselves. — AI systems are rapidly gaining the capabilities to autonomously conduct and advance AI research and development.
• Anthropic Warns Autonomous AI Risks Loss of Human Control — Autonomous AI's rapid self-improvement risks human control loss, necessitating global safety measures.
• AXRP Episode 47 - David Rein on METR Time Horizons — AI's ability to complete long tasks, measured by human completion time, has increased exponentially.
• It's all over in 2 years — Exponential AI growth is driving a societal phase change, rendering many human intellectual contributions obsolete while preserving creative and social domains.
• Dario Amodei (Anthropic) Drops ATOMIC BOMBSHELL at Davos! — AI is rapidly approaching recursive self-improvement, driven by enhanced coding, mathematical intuition, and energetic efficiency.
• AI will help make a Nobel prize-winning discovery within a year, says Anthropic co-founder — Anthropic co-founder Jack Clark forecasts rapid AI progress, including Nobel-level discoveries and autonomous AI, alongside significant existential risks.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Wes Roth.