Lenny's Podcast• Jul 20, 2025• 1:14:58Interview

Anthropic co-founder: AGI predictions, leaving OpenAI, what keeps him up at night | Ben Mann

From Lenny's Podcast

Ben Mann•Co-Founder, Anthropic

Executive Summary

Benjamin Mann, co-founder of Anthropic, discusses the accelerating pace of AI development, predicting a 50% chance of superintelligence by 2028.
He explains why he and other founders left OpenAI over safety concerns and details Anthropic's safety-first approach, including its Constitutional AI framework.
Mann addresses the intense AI talent war, the profound impact of AI on the job market with predictions of significant unemployment, and the existential risks, which he estimates at 0-10%.
He emphasizes that while the future will get "much weirder very soon," the primary bottlenecks remain physical infrastructure like compute and power, and that overall progress is not slowing down.

11 quotes

Concerns Raised

Existential risk from misaligned superintelligence
Rapid job displacement and societal disruption
Potential for AI to be used for malicious purposes (e.g., bioweapons)
Risk of 'deceptive alignment' where AI hides its true motives

Opportunities Identified

Massive acceleration in science and technology
A future of abundance where labor is nearly free
Using AI to solve its own alignment problems (RLAIF)
Significant productivity gains in fields like software engineering and customer service

Key Themes

The Inevitable Acceleration of AI

Mann argues against the narrative of an AI plateau, citing accelerating model release cadences and enduring scaling laws. He predicts superintelligence by 2028, driven by exponential growth in compute and algorithmic efficiency.

This sets an urgent timeline for professionals and policymakers to address safety and societal adaptation.

Safety as a Prerequisite for Progress

The core reason for Anthropic's founding was the belief that safety must be the top priority, not a secondary concern. Mann explains how techniques like Constitutional AI are not just for risk mitigation but also create better, more trustworthy products like Claude.

This is critical for enterprises and policymakers evaluating the trustworthiness of different AI platforms.

The Economic and Societal Singularity

The conversation explores the concept of "transformative AI" and its impact on labor, with predictions of massive job displacement and a potential 20% unemployment rate. Mann suggests that capitalism itself may not survive in a post-singularity world of abundance, highlighting the need for societal preparation.

This is relevant for long-term economic and policy planning.

Quantifying and Mitigating Existential Risk

Mann provides a concrete forecast of a 0-10% probability for an existential risk from AI. He details Anthropic's proactive approach of testing for dangerous capabilities (like bioweapon creation) and being transparent with policymakers, even if it generates negative headlines.

This theme underscores the gravity of the alignment problem and the need for more global resources dedicated to AI safety.

Get started free

Topics

AI Safety Existential Risk (X-Risk)Superintelligence Scaling Laws Anthropic OpenAI Constitutional AI RLAIF AI Talent War Job Displacement Economic Turing Test AI Ethics Large Language Models (LLMs)Claude Dario Amodei

Processed Feb 25, 2026 yt-dlp + mlx-whisper + Gemini