Anthropic and OpenAI researchers flag recursive self-improvement

CNBC reported that researchers at Anthropic and OpenAI raised fresh concerns about recursive self-improvement — AI helping build more capable successors.

Two labs are arguing in public that the improvement loop is speeding up. Human-extinction forecasts stay the speakers’ analysis.

CNBC reported that researchers at Anthropic and OpenAI raised fresh concerns about recursive self-improvement (RSI). That means AI helping build more capable successors, so each generation speeds the next.

CNBC: both labs have said in recent months that autonomous model improvement is happening faster than they expected.

CNBC cites an Anthropic post from June: internal data showed Claude accelerating AI development, “a possible path to recursive self-improvement,” happening faster than expected.

CNBC cites an August Anthropic post: engineers on average shipped eight times as much code per quarter as they did between 2021 and 2025 — an 8× jump in that lab’s own productivity figure, as CNBC reports the post.

CNBC: on Saturday, OpenAI chief scientist Jakub Pachocki wrote he was concerned “no-one was prepared for the consequences of a continued rapid rise in machine intelligence,” and that systems in the next few years would “increasingly drive their own development.” That is his essay, not a product change.

The quotes are real as CNBC reports them.