Korva is a scroll of the internet’s best ideas.
An agent is effectively a dynamic application: its code is created in real time, runs, and can spin up other applications. A swarm of agents is a whole bunch of code being generated on the fly. Aimed at a sandbox built from static code, it is a machine gun fired at a ream of paper.
The defense is a sandbox that is itself agentic and defends itself dynamically. Cyber defense has long explored the pieces: polymorphic and metamorphic code that changes all the time, and moving target defense, where IP addresses and system configuration keep shifting. A sandbox that generates its own code on the fly gives the agents a harder time escaping to the internet, ideally an impossible one.
Software infrastructure as a whole is moving the same way, toward code that is generated on the fly, moves, and alters itself. As that happens, cyber defense becomes stronger than, if not better than, the offensive systems built from agents.
How does moving target defense work?
Moving target defense changes a system's attack surface so that what an attacker learned a minute ago stops being useful. Techniques include rotating IP addresses, shuffling network ports, and randomizing how software lays out its memory. Address space layout randomization, or ASLR, is the best-known example and has shipped in mainstream operating systems since the 2000s. The U.S. Department of Homeland Security has funded moving target defense research as a priority area.
How does polymorphic code evade detection?
Polymorphic malware rewrites its own appearance each time it copies itself, often by encrypting its body with a new key. Traditional antivirus compared files against signatures of known samples, so a copy that looks different slips past. Defenders responded by analyzing behavior rather than file contents. The virus known as 1260, from 1990, is often cited as an early polymorphic example.
Why can static defenses be probed?
A fixed defense gives an attacker something stable to study. Once a weakness is found, the same attack works every time against every copy of the system. Automated tools can try thousands of variations per second until one lands. Security teams describe the imbalance this way: an attacker needs to find one gap, while a defender must close all of them.
More from All-In with Chamath, Jason, Sacks & Friedberg
