Artificial intelligence (AI) has, in one form or another, dominated discussions across technology, finance, government, and society for months if not years at this point. Efficient paperclip production thought experiments aside, early November 2026 saw the emergence of powerful, potentially profound changes in the AI ecosystem. While it is “early,” such trends are not only worth watching, but analyzing as well.
Most direct in AI impacts were the combination of stock market meltdowns and the electoral success of data center opposition in reframing the economic landscape. Along with continuing unease of a potential “bubble” in AI investment that may soon pop (or perhaps is already doing so), these events when combined with upcoming frontier model company public listings paint an unsettling picture. These items, from their impact on 401ks to perhaps unsettling a frothy (but potentially fragile) economy largely built off the valuations of a handful of companies, will get the most mainstream attention and attract the most worry.
They also risk the continuing, rapid development of AI technology, placing its benefits (and there are benefits) further over the horizon due to extreme market correction. Society may find itself in a similar position as the post “dotcom” boom period, where significant investments were made (then in fiber and similar, now in datacenters and compute) that then lie fallow for years due to financial collapse. In this period, some parties may benefit (such as academics or potentially threat actors, as discussed below), but otherwise progress crawls to a halt while organizations recover from the boom-bust cycle.
In more specialized circles, but with potentially greater concern, the rapid weaponization of AI to facilitate malicious cyber operations dominates discussions if not quite headlines. While the US frontier labs “escaping” their masters periodically has perhaps generated the most public attention, the emergence of on-premise, open weight models free of guardrails using frameworks such as ARTEX, on top of existing concerns on vulnerability discovery and weaponization, mean the threat landscape has fundamentally shifted. Multiple, prominent vendors have documented widespread use of AI-driven agentic frameworks for malicious cyber activity while the “vulnpocalypse” continues to unfold in tandem. The result is a world where threat actors simply overwhelm defenders, leading to an exhaustion of resources and incident response capacity.
The first item, the potential collapse of the commercial AI sector (and much economic activity along with it) represents one type of disaster, while the rapid weaponization and offensive application of AI another. Yet the two are linked in an important way: the former will likely delay or otherwise impede the development of AI for defensive and other uses in material ways, while the latter (due to efficacy and profitability) continues unimpeded. Perversely, the weaponization case may even accelerate as adversaries are gifted with cheaper hardware and compute prices due to an AI “bust” in the broader marketplace.
What emerges is a “worst of all possible worlds” scenario for information security, where the harms of AI continue to advance without pause or hesitation, yet the potential to improve defenses through beneficial applications of the same technology whither as markets implode and investments shrink. While it is important to caveat the sense of doom conveyed here with an admission that AI is not unlocking any profoundly unique, concerning, or even especially stealthy malicious tradecraft, it is necessary to accept that the volume and velocity of activity, even if primitive or noisy, has exploded.
If these trends continue on their current trajectory (admittedly a straight line projection, which we should all treat with caution), defenders and network owners will find themselves profoundly disadvantaged. All of the international relations and war studies PhD’s that crowed about offense-defense asymmetry in cyber operations will find themselves finally vindicated thanks to an agentic “Zerg rush” on the part of hackers. Quite simply, life is hard right now in the network defense space—and it is likely to get harder still in the immediate and near-term future.
If the possibilities of defensive AI applications, driven by commercial offerings and the ability to leverage frontier models, evaporate along with share prices and valuations, defenders will need to seek alternative strategies to stem the offensive tide. Although potentially lost in the rush towards the current environment of AI-related buzzwords, terms with a slightly older vintage that are now overlooked, such as “zero trust architecture” and “least privilege” begin to make tremendous amounts of sense in a rapidly weaponized, hostile ecosystem.
Defense will need to abandon a competition focused on speed and scale, with notional agentic defensive workflows and similar, for the foreseeable future, and instead look into reshaping the landscape in defense’s favor. Frequently overlooked and not leveraged to its total potential is the simple fact that, in most cases, defense operates on a network which it also owns, configures, and deploys. Adversaries must deal with and adapt to those decisions, giving defense the ability to architect environments in a way that disadvantages offense and can play to defender strengths. Recognizing this and embracing more restrictive and secure network architectures, as well as more robust (if inconvenient) operating policies and procedures can work to shrink the areas through which adversaries can operate, and buy critical time and space for defenders to detect, respond, and mitigate intrusions.
Examples range from supposedly basic security hygiene decisions, such as not exposing device or application management interfaces to the external internet, to more advanced items like dynamic network segmentation. The former should be implemented immediately, while the latter are objectives for longer-term investment and roll-out. But in both cases, defenders must adopt a posture of preparing and shaping the environment instead of simply applying more bandages to that which currently exists. What was at best somewhat effective in the “beforetimes” is simply incapable of matching the AI-driven swarm of the present.
Unfortunately, these solutions are also hard and expensive to deploy. For a multinational conglomerate with tens of thousands of users, hundreds of thousands of hosts, and millions of accounts and services, “blowing it up and starting over” is neither helpful nor actionable advice. Yet, something approximating such an extreme solution will be increasingly necessary as organizations come to the realization that there is no cavalry coming to their aid, and that the light at the end of the tunnel might very well be a train. Saying “this is impossible” will not solve the problem, while waiting for some deus ex machina only places entities on a perpetual backfoot relative to attackers.
Hubris, greed, and other human frailties look increasingly like they will delay an AI-driven commercial boom, and with it the possibilities to radically enhance defensive actions and response. Yet adversaries have already unlocked sufficient value from available advances that offensive, malicious uses of AI. Such offensive applications are here, right now, and will not go away soon. To meet what appears to be an unfolding crisis, defenders and network owners will need to “go back to basics” and rethink just what is necessary to operate a network that can maintain availability, integrity, and confidentiality to the appropriate degrees for the organization’s mission. Hand-wavvy new technology may assist with this (if the companies developing such solutions don’t starve of VC funding that no longer exists post-crash), but in the right now defenders need to dust off older playbooks, and retake and remake the field of play to their advantage.
0 Comments