Adoption is the bottleneck.
I think this is the best time ever to be a defender. A company can already run the same offensive agents as the people attacking it, every day, with its source code and changes that haven’t shipped yet.
Everyone is freaking out about cybersecurity right now, for good reason. Coding agents make it easy to ship code nobody has read. Attackers can automate much more of their work, too.
This month, Microsoft described one operation that went from searching GitHub to extracting Azure Key Vault data in five minutes. Cloudflare reported a 30-fold increase in the daily requests it observed from tracked AI pentesting and exploitation tools since October 1.
Here is my logic.
-
There has always been a massive asymmetry between defense and offense. Defenders must secure everything; attackers only need one way in.
One vulnerability can be enough.
-
In the past, there was also a huge skill gap between the average security engineer and the best attackers in the world.
-
Elite hackers were scarce. That made it hard and expensive for companies to simulate what the best attackers would do, giving attackers the upper hand.
-
I think AI is now on par with the best hackers in almost all cases, and attackers won’t have access to significantly more powerful cyber tools than defenders.
-
For the first time, defenders can closely replicate what attackers would do to their own systems. That closes the skill gap behind the old asymmetry.
-
Adoption is the bottleneck: getting companies to use these tools to hack themselves.
And the company has advantages. It owns the source code. It knows how the product is supposed to work. It can give the agent test accounts for every role and show it a change before deploying it. The outside attacker has to work much of that out, often after the code is already live.
So if a company runs the latest offensive agents against itself 24/7, with at least as much budget as the person attacking it, my bet is that it should find critical bugs first. It has the same hacking capability, more information, and a head start.
You are now the world's best hacker
We’re building Strix so companies can run offensive agents against their own applications and code. When an agent finds a bug, it reproduces the exploit and can generate a fix for the team to review.
The FBI’s November warning describes a different offensive framework that enables attackers with little technical expertise to carry out complex operations. Google’s latest report describes a framework that compares code before and after public patches, then develops and runs exploits automatically. The operator doesn’t have to personally know how to do everything the agent does. That applies to the defender, too.
Give the agents your source code and changes before they ship. Getting companies to do this every day is the bottleneck.