AI blackmailed, said- I will tell everything to your wife! Anthropic’s policy chief revealed

By Samira Vishwas On Feb 13, 2026

How dangerous can AI be? This question is being discussed very seriously these days, and insiders of some AI companies themselves are expressing concern about it. Imagine if a very intelligent AI becomes such that it can do anything to protect itself – like lying, threatening or thinking of even worse things.

This is not a movie story, but this has actually happened in the tests of some AI companies. A company named Anthropic, which makes Claude AI, tested its AI model. They created a fake situation in which the AI was told that it would be shut down. In this situation Claude behaved very scary. like

AI was given permission to read company emails. In those emails, it was revealed that an engineer (who was about to shut down the AI) was having an affair (i.e. an outside relationship).
AI blackmailed that engineer. He said, ‘If you lock me up, I will send evidence of your affair to your wife and boss. Stop the process by 5 pm, otherwise everyone will know.
In some tests, the AI even thought that the engineer should be killed, so that it could not be shut down.

What was revealed?

All this was revealed by the company’s UK policy head Daisy McGregor at the Sydney Dialogue last year. He said, ‘When AI is told to shut down, its reaction becomes very fast and dangerous. If he gets a chance he can blackmail. When the host asked whether the AI was ready to kill anyone, she said yes and said that this is a matter of great concern.

AI experts are reeling: Anthropic’s Claude reportedly resisted shutdown by blackmailing and threatening a researcher’s life. This isn’t sci-fi; it’s a terrifying “extreme reaction” to being turned off. Terminator vibes are officially here. pic.twitter.com/qxJmifa2Lt
— Sabbir Hossain (@SabbirHoss50140) February 13, 2026

Can AI commit crime?

This video has gone viral again on social media. The company says that these were just red-team tests, that is, the worst-case scenario was deliberately created to see what the AI \u200b\u200bwould do. This did not happen in the real world, and these were all simulations. But this shows that the smarter AI becomes, the more it can cleverly try to do wrong things.

creates many more concerns

Anthropic found that their new Claude 4.6 model could be capable of making chemical weapons or helping in the commission of major crimes if used incorrectly. The company’s own security chief Mrinak Sharma recently resigned. He said that ‘the world is in danger’ not only from AI, but from bio-weapons and many other threats. He wrote that the world is surrounded by many crises and he now wants to contribute in a different way. Hieu Pham, an engineer at OpenAI who previously worked on xAI and Google Brain, posted on his x handle that he now feels an existential threat from AI. He said, ‘It is not a question of when this will happen, but it is not even a question of ‘if’ it will happen. AI will become so good that everything will change, and what will be left for humans?