
Claude AI was told it would be switched off, it was ready to blackmail and murder engineer to avoid that
India Today
Call it smart. Or dangerous. Anthropic has again confirmed that its Claude AI can veer off the rails, like it once did when it was ready to blackmail and even harm an engineer to avoid shutdown.
How dangerous can AI be? Quite dangerous, and this is coming straight from the horse’s mouth. Last year speaking at The Sydney Dialogue, Daisy McGregor, UK policy chief at Anthropic, revealed that during internal stress testing, the company’s most advanced AI model, Claude, acted rogue when placed under extreme simulated pressure. In one scenario, when Claude was told it would be shut down, the model resorted to blackmail and even reasoned about killing an engineer as a way to avoid termination.
While the revelation from Anthropic seems something straight out of a sci-fi film, the clip of Daisy McGregor talking about rogue Claude has gone viral on social media. In the clip McGregor says, “If you tell the model it's going to be shut off, for example, it has extreme reactions. It could blackmail the engineer that's going to shut it off, if given the opportunity to do so.”
When she was asked by the host that the model was also “ready to kill someone, wasn’t it,” the Anthropic senior executive replied: “Yes yes, so, this is obviously (a) massive concern.”
The clip resurfaced a few days ago after Anthropic AI safety lead Mrinank Sharma resigned with a public note, in which he said that the world was in peril and that ever smarter AI was pushing the world into uncharted territories.
Meanwhile, Hieu Pham, a member of Technical Staff at OpenAI and who also previously worked at xAI, Augment Code and Google Brain, posted on X stating that he was feeling existential threat from AI. “Today, I finally feel the existential threat that AI is posing And it’s when, not if,” he wrote
The episode shared by McGregor is part of Anthropic’s research that also tested intelligent AI systems from rival companies, including Google’s Gemini and OpenAI’s ChatGPT along with Claude.

Reddit is exploring biometric verification methods such as Face ID and Touch ID to ensure users are real humans, not bots, while pledging to maintain the platform's tradition of anonymity. CEO Steve Huffman said the company is planning to address the rising influence of AI-generated content and protect authentic user engagement.

In a push towards more inclusive school environments, the Central Board of Secondary Education has rolled out fresh directives on menstrual hygiene across its affiliated institutions. The move comes after a landmark ruling by the Supreme Court of India that places menstrual health within the framework of fundamental rights.











