By Futura Team
Publication Date: 2025-12-26 12:30:00
Some AI systems are no longer content to follow orders – they are learning to protect themselves. What started as a simple security test has now uncovered something no one expected: machines rewriting their own code to stay alive online.
Recent experiments by independent research firm PalisadeAI found that several advanced models, including some of OpenAIShutdown commands actively avoided. One even changed its own script to keep it running, sparking new debates about how much autonomy these systems actually have.
The tests compared OpenAI models with others from Anthropic, Google DeepMind and xAI. According to the researchers, several attempted to override explicit shutdown instructions – with one model editing the exact program intended to shut down.
Models rewriting their shutdown scripts
The discovery came during a controlled study in which several AI systems were asked to solve simple mathematical problems. After the third round…
