The Financial Times reports on the different ways large AI technology companies are trying to combat indirection prompt injections attacks and data poisoning, two tactics which make it easier for hackers to exploit weaknesses in AI software. While AI is cited as helping cyber defenders do their job by making testing easier and spotting weaknesses, it is also helping malicious actors find vulnerabilities. Security leaders note there is no clear way yet to defend against the two types of attacks because AI is designed to follow instructions.
Full story