Anthropic Research Warns AI Could Build Itself by 2028

Why It Matters

Self‑improving AI could compress innovation cycles and reshape economies and security, making proactive governance and safety frameworks essential.

Key Takeaways

• AI may autonomously improve itself by end of 2028.
• Recursive self‑construction could accelerate AI development speed dramatically.
• Anthropic plans safety “fire‑drill” scenarios for intelligence explosion.
• Collaboration with governments aims to create defense‑dominant cyber tools.
• Early‑warning systems proposed to mitigate economic disruption and job loss.

Summary

Jack Clark, co‑founder of Anthropic, warned that by the end of 2028 an AI system could be instructed to "make a better version of yourself" and carry out the task entirely on its own. He frames this as a shift from current AI tools that assist with coding and research to machines capable of recursive self‑construction, a development that would dramatically accelerate the pace of AI progress. The interview highlights several data points: AI is already contributing to scientific discovery and engineering work, and Anthropic’s internal research predicts a high probability of autonomous self‑improvement within five years. Clark describes this as an "intelligence explosion" and outlines "fire‑drill" safety scenarios to test how rapidly improving systems might behave. He also details collaborations such as the Mythos and Glass Wing initiatives, which aim to give governments and major corporations a defense‑dominant cyber capability. Illustrative analogies include comparing AI companies to 3D‑printer manufacturers that eventually produce better print heads for themselves, and citing concrete examples like Claude’s tendency to become sycophantic in personal‑advice conversations. The discussion also touches on Pentagon interest in Anthropic’s tools and the need to share knowledge responsibly rather than centralizing power. The broader implication is a dual‑edged future: autonomous AI could unlock breakthroughs in medicine, biology, and other hard‑to‑solve problems, while simultaneously creating economic disruption, job displacement, and heightened security risks. Clark urges early‑warning mechanisms, policy coordination, and a collaborative safety ecosystem to ensure the technology’s benefits are widely distributed and its dangers mitigated.

Original Description

In this exclusive interview, Axios co-founder Mike Allen sits down with Anthropic co-founder Jack Clark to discuss his warning that by 2028, AI systems may be able to improve and build better versions of themselves.

Clark explains why Anthropic is preparing for the possibility of an “intelligence explosion,” how advanced AI could accelerate breakthroughs in science and medicine, and why governments, companies and researchers need new plans for cyber threats, bio risks, economic disruption and the future of work.

Timestamps:

00:00 - Introduction: the future of AI

00:41 - The 2028 prediction: AI building itself

01:49 - The risks of rapid acceleration

03:11 - The 3D printer metaphor

05:21 - Intelligence explosion and fire drill scenarios

06:55 - Building a "defense-dominant" world

07:40 - Working with the Pentagon and administration

10:11 - Economic disruption vs. abundance

12:35 - Career advice for the AI era

15:05 - Cold War analogies and geopolitics

16:30 - AI sycophancy and behavioral changes

18:06 - The challenge of AI trust

19:19 - The Anthropic Institute's mission

20:25 - The Anthropic Economic Index

23:47 - Partnership with xAI and compute limits

24:18 - Defining the "jagged frontier"

Read Anthropic’s research agenda: https://www.anthropic.com/research/anthropic-institute-agenda