Are We About to Lose Control of AI? | AI Reality Check

June 11, 2026

AI Summary

5 min read

Anthropic’s recent report, “When AI Builds Itself,” arrived with a dramatic animation of machines replicating like cells in a petri dish and a grim warning: if AI begins to improve itself recursively, humans might lose control. The report’s own language is dire, stating that “full recursive self-improvement… could come sooner than most institutions are prepared for.” But when pressed on whether a global pause would help, Anthropic essentially shrugged, saying a slowdown would only help the least cautious actors catch up, so they have no choice but to continue at full speed. The message is clear: we are hurtling toward a potential loss of control, and there is nothing to be done about it except publish solemn reports.

Cal Newport, in this AI Reality Check episode of Deep Questions, argues that this narrative is not just unhelpful—it is wrong. He walks through the three core charts from the Anthropic report to show that the data does not support the doomsday conclusion.

What the Charts Actually Show

Continue reading the full summary in the app — free to try.

Read Full Summary →

Free • No credit card required

Listen to Audio Summary Open in App

What you'll learn

1 (00:00) **Anthropic’s "When AI Builds Itself" Report** - Cal introduces the report’s alarming claims about recursive self-improvement and losing control of AI.
2 (03:02) **Core Data from the Report: Code Productivity** - First chart shows a surge in lines of code per engineer after AI coding tools were introduced in late 2025.
3 (04:14) **Core Data: Claude Code Session Success Rate** - Second chart shows AI solving harder coding problems, with success rates jumping from ~20% to ~70% in 2026.
4 (06:03) **Core Data: Where Researchers Went Wrong** - Third chart shows AI now outperforming human programmers at diagnosing coding problems 59-64% of the time, up from ~50%.
5 (07:54) **Cal’s Verdict: These Fears Are Not Justified** - The data only shows that AI coding tools have gotten good at software development tasks, not that recursive self-improvement is imminent.
6 (09:42) **Reason #1: Faster Software Development ≠ Smarter AI** - The bottleneck to AI breakthroughs is scientific ideas, not coding speed.
7 (12:28) **Reason #2: These Tools Are Completely Controllable** - The coding harness is a human-written, deterministic program; only the LLM output is unpredictable.

+ Full timestamped outline available in the app

Show Notes

Cal Newport takes a critical look at recent AI News.

Video from today’s episode: youtube.com/calnewportmedia

(0:00) Are we about to lose control of AI?

(3:04) How much should we be afraid of recursive self improvement?

(7:52) Are these fears justified?

(9:36) Faster software development doesn’t equal smarter AI

(12:25) These tools are completely controllable

(16:52) Taking a step back

Links:

Buy Cal’s latest book, “Slow Productivity” at www.calnewport.com/slow

https://www.anthropic.com/institute/recursive-self-improvement

Thanks to Jesse Miller for production and mastering and Nate Mechler for research and newsletter.

Learn more about your ad choices. Visit podcastchoices.com/adchoices