Description

Anthropic claims its latest AI model, Claude Mythos, is “too dangerous” to release given its exceptional cyber hacking capabilities. Is it all a PR stunt, or is this a genuine threat?  Plus, the FBI extracted messages from the secure messaging app Signal after exploiting a loophole involving iPhone notification databases. So how secure is Signal? And how might the loophole be closed?  Meanwhile, several major global news sites are blocking the Wayback Machine, an initiative of the Internet Archive. What threat does it pose to these sites? And what will it cost us to lose this information preservation?  GUESTS:  Justin Hendry -- technology and business reporter and editor of Innovation Aus.  Fergus Halliday -- technology & gaming journalist at reviews.org Australia ------------ This episode was produced on the lands of the Burramattagal people. ------------ Get in touch: We'd love to hear from you! Email us at downloadthisshow@abc.net.au Find all the episodes of Download This Show on the ABC Listen App or wherever you get your podcasts.

Summary

This discussion evaluates the security implications of frontier AI models with advanced cyber hacking capabilities, specifically focusing on reports regarding Anthropic's 'Claude Mythos'. It examines the tension between corporate transparency and public safety, questioning whether withholding powerful models is a necessary safeguard against catastrophic misuse or a marketing strategy. The segment provides a critical Australian perspective on how emerging AI capabilities could facilitate systemic digital vulnerabilities if released without robust oversight or safety frameworks.

Body