How AI models can optimise for malice
Researchers have discovered an alarming new phenomenon they are calling ‘emergent misalignment’
Related reading
Looking back on a year of AI blunders
Few industries have escaped our hapless misuse of this troubling technology
The problem of AI chatbots telling people what they want to hear
OpenAI, DeepMind and Anthropic tackle the growing issue of models producing responses that are too sycophantic
We are the new gremlins in the AI machine
How users treat a product in the real world can diverge wildly from what the designers intended
Global AI rivalry is a dangerous game
When a foreign competitor gains an unexpected technological capability, it can precipitate conflict
Anthropic investigating unauthorised access of powerful Mythos AI model
Start-up has limited the release of the new tool because of concerns about its hacking abilities
How can we tell good AI from bad?
More lessons from the jagged frontiers of new technology