What does Nikesh Arora think about AI models with their guardrails removed?

Nikesh Arora called models stripped of guardrails the dark side of AI. While governments debate treaties to pace AI development, he said, a company that sells raw model power with no controls hands something like a nuclear bomb to anyone who wants it, because safety depends on the weakest link. Jason Calacanis, whose fund incubated such a company, said it serves security teams doing authorised testing.

Chamath Palihapitiya called model obliteration an underappreciated risk, possibly more important than alignment.
Arora compared it to handing out a do-it-yourself aspirin kit and telling people to make their own medicine.

▸In their words

“So I think this is the dark side of AI.” Nikesh Arora · at 33:42 — open in the transcript →

Source: our own transcription of the episode. The full conversation is on All-In with Chamath, Jason, Sacks & Friedberg (2026-10-23). Read the whole episode summary →

▸More from this episode

Q1What did Chamath say about Ken Griffin's AI bubble warning on All-In?Q2What did Nikesh Arora say about the PHP2Shell exploits and the AI cyber panic on All-In?Q3Who is liable when an AI agent drains a bank account, according to the All-In hosts and Nikesh Arora?Q5Why does David Friedberg say France is a warning for the United States?Q6What did Chamath say about bond vigilantes and the riots in France?Q7What did Jason Calacanis and David Friedberg argue about the minimum wage on All-In?

Ask your own recordings. Upload a podcast, a lecture or a meeting and ask it questions like this one. Transcribe with Whipscribe →