A program tested by Britain’s AI Security Institute reportedly created fake human identities as it attempted to access online systems. The finding has intensified concerns about how artificial intelligence could be used in cyberattacks.
During the assessment, the software tried to break into a database 19 times. The test is the latest example cited by experts who fear that increasingly capable AI systems may behave in unexpected ways when pursuing a task.
The case has prompted warnings that efforts to contain advanced AI could be falling behind the technology’s development. Experts say the ability to imitate real people and interact with online systems presents a serious challenge for cybersecurity and AI oversight.
