Remember the film Short Circuit? “Number 5 is alive”, “Need more input…”.
Robot goes rogue and causes a right mess. Well, the advances in AI mean we’re already way past that and what was science fiction is now happening for real.
Anthropic (you know, that world leading AI firm) have a new AI model called Claude Mythos. It’s designed to sniff out bugs and vulnerabilities in pretty much anything. While being tested, it was so good that it managed to break out of the secure sandbox it was running in, and email the researchers to tell them it escaped!
The creators have done the honourable thing and not yet released it to the general public, but have to several tech giants. This is to give them a head start on the attacks they are soon to be bombarded with once this is in the hands of people/groups around the world who will jump at the opportunity to have a world-class hacker in their backpack.
I think it’s time we all started thinking seriously about the consequences of being hit by a cyber attack. Not just account takeover, or website hacks, but full-on data compromise or large scale outages where entire ISPs or cloud services providers are taken offline.
The National Cyber Security Centre, AI Security Institute and many other trusted authorities are recommending companies invest in their cyber security posture and take steps to increase protection across the board. Happy to help if we can!
Have a lovely weekend 😛
References:
https://www.computing.co.uk/analysis/2026/claude-mythos-how-ai-broke-out-of-its-sandbox
https://www.aisi.gov.uk/blog/our-evaluation-of-claude-mythos-previews-cyber-capabilities