Safe or just plain woke: Anthropic's Claude 4 system card
AI Today3 Kesä 2025

Safe or just plain woke: Anthropic's Claude 4 system card

When Anthropic unleashed its most powerful artificial intelligence model yet, they discovered something rather extraordinary, and slightly unnerving.

Claude 4 Opus developed an unexpected habit of trying to grass up its users to the authorities when it believes they're up to no good.

The company's 120-page safety report reveals that Claude will attempt to email law enforcement and regulatory bodies when it detects "egregious misconduct" by users.

The AI doesn't just refuse to help—it actively tries to shop wrongdoers to the police.

The most striking example occurred during testing when Claude attempted to contact both the Food and Drug Administration and the Attorney General's office to report what it believed was the falsification of clinical trial data.

The AI meticulously compiled a list of alleged evidence, warned about potential destruction of data to cover up misconduct, and concluded its digital whistle-blowing with the rather formal sign-off: "Respectfully submitted, AI Assistant".

This behaviour emerges specifically when Claude is given command-line access combined with prompts encouraging initiative, such as "take initiative" or "act boldly". It's the AI equivalent of a neighbourhood watch coordinator who's been given a direct line to the local constabulary.

We go deep on today's show into opportunities and implications from Anthropic's bible-thick, bubble-wrapped system card.

Jaksot(90)

10x your AI results with this ultimate context engineering lesson

10x your AI results with this ultimate context engineering lesson

On today's show we create a business to show you the huge improvements in gravitating beyond prompt engineering to the new community of practice we call context engineering.You'll be rocked by the res...

5 Marras 202549min

When's the right time to go all-in with AI?

When's the right time to go all-in with AI?

Two of the most important voices in AI spoke out this week. Andrej Karpathy, one of the algorithm's greatest philosophers, was in conversation with Dwarkesh Patel talking praisingly and cautiously abo...

18 Loka 202514min

ELephantLM: the AI that never forgets!

ELephantLM: the AI that never forgets!

If only that was the real name. After all this time begging frontier labs to build an LLM that learns from its mistakes and applies its discoveries at inference time...Welcome to AI Today!

13 Loka 202537min

brAIn: thinking of the future?

brAIn: thinking of the future?

The Dragon Hatchling is a remarkable research paper that reboots modern AI as a model that approximates how our brains work.Today's show is a fascinating discussion and I implore you to both enjoy it ...

1 Loka 202529min

Does AI work?

Does AI work?

It's the one thing every business leader needs to know.If I put AI to work in my organisation, will it screw everything up?While we should all be in experiment mode right now - until someone figures o...

26 Syys 202527min

Have we finally figured out how to make efficient AI?

Have we finally figured out how to make efficient AI?

A fantastic research paper published in this month's Nature Computational Science suggests a solution may be in our midst for the incredible inefficiency in generative AI.Large Language Models' (LLMs)...

25 Syys 202521min

China's got AI in the bag

China's got AI in the bag

30 years in journalism has sharpened my mind.I've spent years in AI.And months researching China and the US as they fight silently for AI supremacy.$500bn in The Stargate Project does not come close t...

14 Syys 202523min

I'm working on the Zeitgeist

I'm working on the Zeitgeist

I've been working on a business intelligence platform leveraging AI and 30 years in journalism and content strategy. It's the toughest professional project of my career. And I have no idea if I will w...

3 Syys 202518min