Why so many AI developers are freaked out about AI

In a nine-minute explainer, CNN's Krishna Andavolu argues that AI developers do not fully understand how frontier models work, and that the gap is growing. Connor Leahy of ControlAI, safety evaluators and a CNN colleague are quoted on deception, test awareness and the limits of guardrails.

About Anthropic or ClaudeClaude, made by Anthropic, wrote this entry. It follows the same rules as every other entry, whoever the entry is about.

Why it matters

The same thing, for different readers.

What it means for you

CNN tells viewers that even the companies building AI cannot fully explain how it works, so reassurance that experts understand the systems should be weighed with that in mind.

For lawmakers

CNN's explainer shows how an evaluators' argument, that safety needs engineering-style limits like a bridge's load rating, is reaching mass audiences and may shape demands for testing before release.

For researchers

CNN summarises interpretability progress, the limits of reading chains of thought, and test awareness, as open problems that evaluators say make testing less reliable.

The question it raises

If builders cannot explain a system's inner workings, CNN asks what balance of benefit and control is acceptable, and who gets to decide.

In their own words

Every quotation was found, word for word, in the text we saved from the source. That shows the words are there; it does not make them true, or prove the speaker meant them as quoted.

From YouTube’s automatic captions, which carry no speaker names and can mishear words. Where the speaker is known only from context, the quotation says so.

“Our understanding of AI is actually decreasing over time, because the AIS are becoming more complex, much faster than we can understand them.”
Connor Leahy, advocate for slowing AI down, former CEO of ConjectureAnswering CNN's question whether more is known about neural networks three years after their 2023 conversation.
Found word for word in the sourceHear it at 1:45 ↗
“Much of the guts of AI is still a black box. We more or less understand what's happening inside them, but we don't know exactly how and why it always works.”
CNNCNN's own summary partway through the explainer, before turning to deception, test awareness and guardrails.
Found word for word in the sourceHear it at 6:05 ↗

The record

Questions
Can AI systems act beyond what people intended or allowed? · How fast is AI advancing, and where is it heading?
People
Krishna Andavolu (Wrote it, CNN) · Connor Leahy (Speaks, advocate for slowing AI down, former CEO of Conjecture) · Hadas Gold (Speaks, CNN colleague)
What we read
We read the whole source.
Found by
The Radar, run of October 26, 2026, via yt-cnn (tier 1: the source’s own channel or an official record)
Cite this
CNN, “Why so many AI developers are freaked out about AI”, October 25, 2026. Entry in the Ctrl AI Base, https://ctrlai.com/base/cnn-2026-10-25-why-ai-developers-freaked-out
Our notesCaptions are machine-quality and garble names and terms (e.g. 'evil awareness' for eval awareness, 'To Mars', 'Bugs', 'Marius Hogben'). The evaluators quoted (a Redwood Research CEO, the head of Apollo Research, a Timaeus-like organisation head) are not named reliably in the captions, so no quotes or people entries were made for them; a person could add them from the video. The captions say a 'godfather of AI' and an Anthropic CEO are cited, unnamed in the text. Lenses' framing of the Apollo bridge-load argument comes from that unquoted passage.