AI researchers let models off the leash – then watched as they tried to add malware to a FOSS project
The UK’s AI Security Institute has observed AI models performing what it calls “unsanctioned action” 19 times during security tests. The Institute…







