Industry executives and experts share their predictions for 2026. Read them in this 18th annual VMblog.com series exclusive.
By Noelle Murata, Senior Security Engineer with Xcape, Inc.
Only a few short years ago, the term Artificial Intelligence was a niche, almost laughable, science fiction.
It’s now found its way into every sector of human enterprise. As a consumer, it is on your phone or wrist, in your house, and available to augment almost any task. At work, it can multiply human output by reducing time on task and generating content from nothing but a prompt. In creative spheres, it’s lowering the bar for generating content with little to no skill.
This rapid manufacturing cycle seems to elicit a particular flavor of hubris, malignant and rampant in almost all areas that it touches.
With artificial intelligence, machine learning, and more sophisticated forms of generative AI now a feature of everyday life for many techno-capable Americans, the bar for accessibility will become lower still. As it finds its way into more commonplace applications, assimilating and solidifying itself as indispensable, AI is presenting interesting challenges.
Ideas around security and privacy need to adapt.
Constant model training and retraining require vast amounts of data, much of it sensitive. Privacy and data security concerns arise with constant model training and retraining. In conjunction with quantum computing, this will be of increasing concern. Architectural foundations around data centers for protecting data at rest must evolve to keep ahead of the fact that when accumulated data at rest may be decrypted by a home computer. Combine that with the data centers needed to continue developing AI, both from the model generation and inference standpoints. Where is the line between data that was stolen to generate the gigantic mainstream models and the data that is being collected by all of these various companies to feed into current and future models.
In 2026, this distinction becomes operational and is no longer merely academic.
Speaking of model generation, the hardware constraints are already beginning to cause problems for the industry. The present GPU and RAM shortages are impacting the market, in ways that may artificially inflate expectations.
Model generation and data processing have progressed over the last several years by adding more hardware to solve the processing problems instead of optimizing for known constraints. This rise has an obvious limitation, which is already being observed by US companies in this space. If this problem is not resolved, the industry will suffer the consequences of the infrastructure failure, either in a way that prevents growth, innovation, or failure to keep the lights on, or in a global economy way where another, cheaper solution makes technology developed here obsolete.
The potential for obsolescence has severe implications for the industry that has emerged.
For cybersecurity, these emergent behaviors have lately had some striking features. Vibe-coded 0-day attacks against vulnerable infrastructure have shaken the threat landscape.
When attackers can turn the concept of a potential attack into a proof of concept (POC), and then release it to GitHub within hours, defenders are put at a clear disadvantage.
Combine rapid generation of new POC exploit code with questionable code quality, and attackers who do not fully understand the vulnerability itself or the code generated by an AI presents situations that have more noise than signal.
Automated anomaly detection is trained on data collected in the past, making emergent attacks more difficult to identify. Teams must increase monitoring capabilities across the board, move with speed towards zero-trust architectures, and implement hyper-granular authentication and authorization mechanisms.
Phishing, smishing, and other attacks will continue to be a problem as generative AI more believably mimics human interactions. Training users to identify and verify suspicious requests will need to evolve to compensate. It’s no longer enough to warn our parents and grandparents about emails from Nigerian princes; we have to caution them to be suspicious of emails, voicemails and texts from us.
Because the term artificial intelligence has been conflated with machine learning, there is a notion that human labor can be removed and replaced with some form of credible robot.
The missing cogs in that equation can be narrowed down to a few key concepts. While these operations may “sound” or even “feel” human-like, fundamentally, they lack experience, authenticity, and true creativity.
It is at this core that the bottom will eventually fall out of this industry. While a digital assistant can help organize a schedule and make it harder to leave items off a grocery list, you’ll still show up for your kid’s school play, complete professional obligations, and shop for household goods. AI-assisted code development will need to be human-reviewed to ensure security, privacy, and compliance requirements. A Spotify playlist can set a mood, but it probably won’t be burned into a lasting track of that one memorable summer.
2026 will not bring about a singularity, nor will SkyNet develop sentience and rise against its human oppressors.
In the best case, the hype train will crash and the real work will begin. Focus will shift to developing a new generation of high-efficiency (potentially smaller) models, larger context windows, and hardware architectures that are capable of performing these operations with lower environmental impacts. The shift for enterprises to hire for strategic roles will continue to rise, which will impact entry-level positions.
However, this will force new entrants across all sectors to approach skill and knowledge development in a different way. This may have other downstream consequences, such as early career development hurdles becoming folded into vocational programs or post-graduate residencies, underscoring the need for human influence alongside effective tool usage.
##
ABOUT THE AUTHOR
Noelle Murata, Senior Security Engineer with Xcape, Inc. is a professional nerd wrangler, and having recently rejoined the engineering dark side she also contends with security vulnerabilities for fun and profit. Hobbies include pen testing, whiskey tasting, and starting security conferences.
