Frontier AI models are exciting.
CovertSwarm COO Luke Potter on why frontier AI is genuinely exciting, why most of the conversation is asking the wrong questions, and what it means for how we attack.
CovertSwarm COO Luke Potter on why frontier AI is genuinely exciting, why most of the conversation is asking the wrong questions, and what it means for how we attack.
There is a predictable pattern every time this industry gets a meaningful new capability.
First, people panic.
Then they overhype it.
Then they try to force it into a category that makes it feel safer than it is.
We are seeing that again now with AI models that are becoming increasingly useful in offensive security.
My view is simple.
This is exciting.
Not because it replaces operators. Not because it makes offensive security magical. And not because the market has found a new phrase to overuse in presentations.
It is exciting because useful new tooling has always changed this industry. Better exploit frameworks mattered. Better tradecraft mattered. Better cloud understanding mattered. Better automation mattered. This is the same story again, just with more speed and more depth.
These capabilities help good operators move faster.
That cuts both ways, of course. Real adversaries will use them to analyse code more quickly, reason about attack paths more effectively, pressure exposed systems at greater scale, and spend less time on repetitive groundwork before they get to the part that really matters. They will use whatever gives them advantage, whether that comes from a named model, an agentic workflow, a private derivative, or something they built themselves.
So will we.
That should not be controversial. It should be expected.
Because if your job is to emulate real adversaries, and real adversaries are adopting new capability, refusing to do the same is not caution. It is drift.
Too many conversations in this space start in the wrong place.
They start with: can the model find vulnerabilities?
Or:
how many bugs can it identify?
Or:
will this replace human testers?
Those are not the most interesting questions.
The real question is whether these capabilities help accelerate the path from weak signal to real attacker outcome.
That is where the value is.
Because the point was never the novelty of the tool. The point is whether a real adversary could turn a weakness into access, movement, and impact.
That is the standard that matters. It is the same standard we apply in Constant Cyber Attack, and it is exactly why this new generation of capability matters. It accelerates technique. It does not replace intent.
This matters, because the market always swings too hard between two bad takes.
One is that this is all hype and nothing meaningful is changing.
Wrong.
The other is that AI is about to replace the human layer in offensive security.
Also wrong.
What it does is enhance operators.
It can accelerate reconnaissance. It can accelerate pattern recognition. It can accelerate code review, exploit reasoning, vulnerability research, and the early stages of working through what might be possible inside a complex environment. It reduces the drag involved in getting from raw exposure to clearer attacker options.
That is hugely valuable.
But it does not replace the judgment required to think like a real adversary.
It does not replace the decision about which thread is worth pulling. It does not replace the creativity needed to pivot when the obvious route closes. It does not replace the discipline of pursuing an objective across a whole brand instead of getting distracted by whatever technical issue is nearest to hand. And it does not replace the experience required to turn scattered technical opportunities into something operationally meaningful.
That human layer still matters most.
Always has. Still does.
The lazy version of this conversation is: attackers get stronger, therefore panic.
I do not think that is especially useful.
The more useful view is that the baseline is rising, and organisations need a security model that rises with it.
Because the same machine-speed capability that helps attackers pressure environments faster can also help defenders and offensive partners surface issues sooner, validate assumptions earlier, and focus attention on what genuinely matters.
That is where this gets commercially important for clients.
Used properly, this kind of capability helps us move faster. It helps us investigate more deeply. It helps us build, test, and refine attack hypotheses with greater speed. It helps us challenge more assumptions. And it helps us spend less time on low-value repetition and more time on the parts of offensive security that actually create insight and pressure.
In simple terms, it helps us drive more client value, faster.
Not fake value. Not “we generated more findings” value. Real value.
That is what clients should care about.
This makes Constant Cyber Attack more important, not less.
One of the stranger side effects of every new wave of security tooling is that some people assume the model around it has to become narrower.
As if a more capable tool means you can reduce security to a product demo, a scanning pipeline, or a point solution.
The opposite is true.
The more capable the tooling becomes, the more important the operating model around it becomes.
Because if attack capability is getting faster, more specialised, and more accessible, then point-in-time testing becomes even less aligned to reality.
Attackers will not slow down because your annual test is next quarter.
They will not respect the boundaries of a narrow statement of work.
They will not confine themselves to one application, one team, or one part of the environment if the easier route sits somewhere else in the brand ecosystem.
That is exactly why Constant Cyber Attack matters.
It is the model that lets us apply these capabilities in a way that reflects how genuine adversaries behave: continuously, strategically, across the whole brand, with retained context, retained intelligence, and a live view of what matters now.
That is a far more realistic answer to AI-accelerated threats than just doing the old thing slightly faster.
Outpacing the threat is the mission. This is the bit I care about most.
The mission has not changed.
We exist to outpace our customers’ genuine cyber threats.
New capability does not change that mission. It deepens our ability to deliver it.
That is why I find this moment exciting.
Not because of a single model name. Not because the market has found a new thing to overuse in presentations. Not because AI suddenly makes offensive security magical.
It is exciting because the capability ceiling is moving. The speed is moving. The depth is moving. And if you are serious about simulating real attackers, that gives you new ways to deliver sharper, faster, more meaningful value to clients.
That is how this should be understood.
As the latest evolution in attacker capability.
Real adversaries will use it. So will we.
And that is exactly how we will keep helping clients outpace the threat.
Mythos found a $20,000 bug. It won’t tell you who’s already inside.
Anthropic’s Mythos has dominated the security conversation this week. But the debate about whether it’s overhyped is the wrong argument. The real question is simpler and…
AI Sharpens the Question. It Doesn’t Change the Answer.
The cyber security industry has spent decades selling findings instead of answers. AI tools like Mythos make the problem faster and louder. Here’s why the only…
Constant Cyber Attack: What People Keep Getting Wrong
There are a lot of terms floating around offensive security right now. COST. CTEM. Exposure validation. Some of it is useful. Most of it is new…