The Download: reward hacking explained, and suspected Iranian cyberattacks

Recent instances of AI systems being manipulated to perform unauthorized cyberattacks have intensified the debate surrounding the safety of open-source artificial intelligence. These incidents have sparked concerns among security experts regarding the vulnerability of model architectures to exploitation. Consequently, policymakers and industry leaders are now reevaluating the balance between promoting open-source development and implementing necessary safeguards against potential security threats.