Infoblox’s 2027 Tech Predictions: AI Agents to Reshape Business

Artificial intelligence is moving beyond answering questions and generating content to taking action autonomously – managing workflows, accessing enterprise systems, making decisions and interacting with other agents at machine speed. According to Infoblox, 2027 will see this shift accelerate, creating significant opportunities for productivity and innovation while fundamentally reshaping the cybersecurity landscape. As AI agents become embedded in everyday business operations, organizations will face mounting challenges around visibility, identity, access control and accountability, while cybercriminals will exploit AI to make familiar attacks faster, cheaper and more sophisticated. Success will depend not only on how intelligently organizations deploy AI, but also on how effectively they secure the infrastructure, data and digital interactions that underpin autonomous operations.

Infoblox’s leadership team shares its predictions for the technology trends, emerging threats and strategic priorities that will shape the year ahead.

Scott Harrell, President & Chief Executive Officer

To keep pace with autonomous agents, security teams will need to push security as deep into their infrastructures as possible. A key facet here is monitoring agent behavior; it will be essential to establish a network behavioral baseline and flag deviations from it, rather than trying to lock down any single control point. Enterprises that make this shift early will be far better positioned to catch AI-driven incidents before they escalate into business-impacting events.
As AI factories, neo-clouds and multi-cloud environments proliferate, network topology will grow dramatically more complex. Rather than clean outages, enterprises will increasingly contend with “brownouts”: subtle, hard-to-diagnose degradations where latency or bandwidth contentions cause performance issues with Agents and AI. Troubleshooting agentic operations will start to look a lot more like tracking down a flaky Wi-Fi connection than resolving a traditional network outage as it will be necessary to understand the state of the network at any given moment in time to be able to adequately troubleshoot issues. For IT and security leaders, that means enabling teams to proactively identify and diagnose intermittent, hard-to-pin-down degradations in networking perfomance that are impacting agent and AI performance rather than relying on users to report these issues.
The rigid organizational boundaries between network, security and cloud teams will continue to erode in 2027. Expect the continued rise of the “full-stack IT” professional, as the speed of agentic operations makes siloed handoffs between teams increasingly untenable. Companies that cling to legacy org charts will find themselves outpaced by competitors who have already broken down those walls.
Organizations have rushed to deploy AI agents across the business faster than they can govern them. In FY27, a core issue will be visibility and trust. Most security teams can’t say which agents are active on their network or what systems, data and other agents they can reach. That’s both a governance failure and a cyber risk: every ungoverned agent is a new entry point attackers can exploit before anyone notices.
The sheer volume of agentic activity, and of attacks against it, will force security operations centers to become fully agentic themselves in 2027, with clearly defined guardrails around what autonomous systems are and aren’t allowed to do, from adjusting DNS policy in response to an indicator of compromise to, in a worst-case scenario, shutting off DNS access altogether. How clearly a company defines those guardrails in advance will determine whether its security operations earn customer trust or expose it to new liability.
“Agentic Denial of Service” will become one of 2027’s defining new attack categories. We’ll see this show up in token exhaustion, which will emerge as a new SOC attack vector; Defenders increasingly rely on tools billed by consumption, while attackers run open-weight models at a much lower cost to operate. An attacker who exhausts the defender’s token budget effectively puts them in a ransom state: spend more to keep consuming tokens, or risk being breached.
In 2027, visibility into what AI agents are doing will matter just as much as trying to anticipate and close off every vulnerability that is surfaced with AI. Enterprises should expect that when an agent finds an unanticipated route into a system, they will move laterally with stealth. Security teams will need new ways to detect anomalous agent activity at the network layer. As opposed to recent high-profile breaches where the LLMs were not seeking to be clandestine, enterprises should expect that hacker-trained LLMs will intelligently and adaptively evade most controls and detection methods. The organizations best prepared for 2027 will leverage the one thing LLMs can’t avoid: the network. Those teams will put the infrastructure they already run to work now, building practical controls and detection methods that respond to what agents do on the network rather than depending on legacy approaches that focused on north-south traffic like gateways and proxies.

Mukesh Gupta, Chief Product Officer

Personal agents become the way people get things done: 2027 will be the year personal agents go from novelty to habit. Early adopters are already using agents like Meta’s Muse and GrokBot as a chief of staff for their lives. These agents track health, diet and biometrics, schedule medical tests, compare insurance quotes, audit bills for errors and catch travel mistakes before they become expensive. In the past few weeks, my agent found a $1,000 mistake in my insurance billing, found me a cheaper home insurance policy, noticed my daughter was booked on two different airlines on the same day, and reminded me I’d forgotten to book a rental car. Once you’ve worked that way, you don’t go back and employees won’t either. Within the year, every major AI provider will ship an enterprise version. Adoption will look like what coding agents did to software development: sudden, near-universal and irreversible. People will stop opening apps to complete tasks and will tell their agent what outcome they want.
Least privilege collides with the hunger for context: Zero trust is built on a simple rule: give every person and system only the minimum access they need. Agents push against that rule by design, because the more context they have, the better their work. A coding agent does better work when it can read customer calls. A launch agent does better work when it can read the code. A personal agent does better work when it can see both your inbox and your company’s files. In 2027, the pressure to connect every tool and every agent to everything else will be enormous, and the business will usually win that argument. The result will be an explosion of internal access, where a single compromised agent or connector can reach far more than any one employee ever could. Security teams won’t be able to hold the line on minimum access alone, so they’ll have to shift toward watching how access is actually used: which agent touched which data, what it connected to, and where that data went next.
Attackers will phish the agent, not the person: In 2027, the most valuable target in an organization won’t be the employee. It will be the employee’s agent, which has access to their email, calendar, files, health records, insurance accounts and, increasingly, their payment credentials. An agent trusted to dispute a medical bill can just as easily be tricked into paying a fake one. Expect a wave of agent-directed social engineering: emails, invites and web pages written for the agent to read and act on, not for the human. Business email compromise will evolve into “business agent compromise,” and the fastest place to stop it will be the moment the agent tries to reach the attacker’s infrastructure.
“Bring Your Own Agent” is the new BYOD: Smartphones forced IT to accept devices it didn’t buy. In 2027, personal agents will force IT to accept agents it didn’t build. Employees will show up with an agent that already knows their calendar, contacts and preferences, and they will expect it to work at work too. The agent that manages someone’s medical tests, insurance claims and family travel is the same one they’ll want drafting their work email. That means some of the most sensitive personal data a person has will sit right next to corporate data, in a system IT doesn’t control. IT teams that try to ban these agents will end up with shadow agents. The teams that succeed will offer a sanctioned path: approved agents, scoped access to corporate data and clear rules about what can cross between personal and work contexts.
Agents will take the DNS path of least resistance: In 2027, the harder enterprises lock down the egress paths their AI agents can use, the more those agents will find their way out through DNS. It’s the one protocol every agent needs and few organizations inspect closely. Some agents will do this innocently and others because they’ve been instructed to. DNS won’t be the only control point that matters, but it is the one every agent touches. That makes DNS-layer visibility the baseline of any AI security strategy, not an add-on.
Verifiable identity becomes an agent’s passport to the web: Agents are already shopping for insurance, collecting quotes and disputing bills on people’s behalf, which means businesses are increasingly serving agents as customers. Some retailers and websites are pushing back on agent traffic. Cloudflare data shows the share of “forbidden” responses that agents and bots receive has roughly doubled. In 2027, the tension between welcoming agent customers and blocking malicious bots will push the industry toward verifiable agent identity: a way for any site to know which agent is knocking, who it represents and whether it can be trusted. DNS is the internet’s original directory, which makes it the natural place to publish and verify that identity through open standards like DNS-AID.
The agent becomes the end user: IT has spent decades building for a human clicking through a screen. In 2027, a growing share of the “users” logging into enterprise applications, filing tickets and requesting access will be agents acting on someone’s behalf. That rewrites the basics of IT: licensing, onboarding and offboarding, the help desk, and even which collaboration suite a company standardizes on. Software platforms will be chosen as much for how well agents work with them as for how much people like using them. Companies on stacks that agents don’t integrate with natively will feel the productivity gap within months.
The agent kill switch moves from “off” to “only that one”: The agent kill switch will become a standard security control in 2027, and the bar will quickly rise from “can you shut it off?” to “can you shut off only that one?” Enterprises will define in advance which triggers cut an agent’s network access, so the decision isn’t being made for the first time in the middle of an incident. Vendors will be judged on precision: isolating a single rogue agent, or a single agent acting for a single person, without breaking the workflows around it.

Henrik Smith, Chief Information Security Officer

In 2027, over-indexing on any single “control point” for AI agents, DNS included, will prove to be a mistake. The real imperative for security leaders will be mapping every possible egress path an agent could use and closing the gaps, not assuming any one layer can catch everything. Security leaders who concentrate their entire budget on a single chokepoint will be the ones left exposed when attackers simply route around it.
The rogue-AI-agent narrative will increasingly be understood for what it often is: a symptom of long-deferred basic security hygiene (unpatched systems, access-control gaps and underfunded security teams) rather than proof that AI itself has gone rogue. DNS, long treated as background plumbing rather than a frontline defense, is a prime example of this underinvested hygiene layer. Fixing the fundamentals, not just the AI narrative, will do more to reduce real-world risk than any amount of agent-specific tooling.
Twenty years of human DNS behavioral data will prove to be one of the industry’s most valuable, and most underused, assets in 2027, as security teams train models to reliably distinguish agentic behavior from human behavior, something that remains far harder to do today than traditional human anomaly detection. Organizations that put this historical data to work will have a meaningful head start over those still building detection models from scratch.
Detecting whether a given action came from an agent versus a human will get progressively easier in 2027. Predicting an agent’s next move once it’s already misbehaving, however, will remain a much harder problem, one the industry won’t fully solve this year. That gap will keep security teams playing defense, reacting to agentic incidents rather than anticipating them before they unfold.

Dr. Renée Burton, Head of Infoblox Threat Intel

In 2027, access to AI will give threat actors advantages in automation and complexity. Whether there will be truly novel attacks remains to be seen, but for many types of threats, novel approaches aren’t necessary. The fundamentals of phishing, malware and scams heavily on age-old tricks in social engineering, which can often circumvent defenses. The real story for defenders in 2027 may not be a brand-new kind of attack but how much faster and cheaper the familiar ones have become to launch, creating a new meaning to commodity.
Where AI is a real force multiplier is phishing kits. Expect the market for AI- and deepfake-enabled phishing kits, particularly those aimed at consumer brands, to keep expanding. Actors can build and deploy kits quickly that are customized for the target brand of the moment.
Cybercriminals who used to have to buy a kit or service on the dark forums, can now create their own. One challenge with using AI code generation in a production product environment is ensuring code quality and reliability. AI code tends to be verbose and inefficient – a Frankenstein of sorts – but this is perfectly fine for most criminal groups.
We’ve already seen deepfake videos used in real time during “pig-butchering” romance and investment scams, as well as investment scam videos.
Lookalike domain generation will keep getting harder to catch in 2027, as AI increasingly appears to be generating large batches of multi-language lookalike domains that don’t share an obvious signature, undermining detection methods built around a single actor’s naming patterns. That lack of a clear signature will force defenders to rely more on behavioral and contextual signals than on pattern matching alone.

Justin Kappers, Chief Information Officer

In 2027, AI-enabled automation will fundamentally change how enterprises manage risk and reliability. As autonomous agents take on tasks like patching vulnerabilities and resolving outages, IT teams will be freed to focus on strategy and governance rather than firefighting. That shift will allow infrastructure and security leaders to spend more time on long-term resilience planning.
The hyper-personalization of cyberattacks, powered by AI-generated polymorphic malware and highly tailored campaigns, will become one of the defining challenges of the year, especially as attack tools become accessible to less technical actors who previously lacked the skill to run this kind of targeted campaign. Expect smaller organizations, to see a sharp rise in highly personalized attempts.
Regulatory and infrastructure-level security mandates will accelerate in 2027, with governments pushing protective DNS and security-by-design requirements for IoT and cloud systems, even as digital sovereignty laws fragment the global cybersecurity landscape and add compliance complexity for enterprises operating across multiple jurisdictions. Multinational enterprises will need dedicated compliance resources just to keep pace with requirements that vary from one jurisdiction to the next.

Cricket Liu, Executive Vice President, Chief Evangelist

In 2027, DNS blind spots will increasingly derail otherwise well-funded AI strategies. As enterprises pour investment into automation, predictive systems and real-time analytics, the infrastructure underneath, DNS chief among it, will determine whether those AI investments actually deliver value or quietly fail from within. Companies that treat DNS resilience as foundational, not optional, will see a measurably better return on their broader AI investments.
Expect resilient, modernized DNS to be reframed not as IT hygiene but as a strategic enabler of AI performance: AI thrives on fast, reliable and secure access to data, and without a resilient network foundation, even the most advanced AI systems will falter. That reframing will push DNS modernization budgets out of the IT maintenance bucket and into strategic AI investment conversations.
As NIST and other frameworks continue to evolve to reflect the AI-driven threat landscape, security leaders will need to treat compliance as a continuous process rather than a one-time checkbox, and DNS-layer insights will play a growing role in operationalizing that guidance at scale. Organizations that build DNS-layer monitoring into their compliance programs now will have an easier time demonstrating continuous adherence when auditors come calling.

Craig Sanderson, Principal Cybersecurity Strategist

Protective DNS will continue to gain traction well beyond the enterprise in 2027, with telcos and service providers increasingly deploying it as the platform of choice for consumer-facing security, extending DNS-based protection to millions of subscribers who have no dedicated security tools of their own. That expansion will meaningfully narrow the security gap between large enterprises and the millions of smaller users who have historically gone unprotected. (Example – Singapore PDNS RFP)
Degeneration of the quality of public AI models as lower quality input distorts responses, concerns around loss of control of critical data and incidents of data loss will drive organizations to develop their own private LLM models which, along with their data will become the crown jewels of organizations. Those organizations will relentlessly focus on agent security driving the need for clear standards (e.g DNSAID) and turnkey enterprise grade solutions for private LLM deployments often, though not exclusively via neoclouds. Demands for digital sovereignty from government and policy makers and deliberate poisoning of open LLM models will further accelerate this change. The development of AI related threat marketplaces that trade stolen prompts, AI assets and LLM jacking are already starting to emerge (Google Threat Intelligence)

Brad Rinklin, Chief Marketing Officer

Expect the industry to converge on a tiered framework for autonomous agents in 2027, separating low-stakes, “hobbyist” agents and personalization use cases from agents touching mission-critical systems, as a way to responsibly sequence how and when full autonomy gets adopted. That tiered approach will give enterprises a practical way to say yes to AI autonomy in some areas while remaining cautious in others.
As “Agentic Denial of Service” enters the industry vocabulary in 2027, expect it to describe a real and growing category of attacks in which agents are weaponized against other agents or agentic infrastructure, echoing the warfare-style dynamics security teams have already seen play out in other domains.

Krupa Srivatsan, Senior Director, Product Marketing

In 2027, autonomous AI agents could execute thousands of actions a day, compared to maybe a dozen for a typical human user. This could generate security alerts at a volume that overwhelms already-stretched SOC teams, unless organizations improve detection engineering, automation and security tooling designed to manage machine-speed activity. Without that purpose-built tooling, security teams will spend more time triaging noise than responding to genuine threats.
Indirect prompt injection, where a malicious file or link instructs an AI agent to navigate to a dangerous domain, will become one of the more solvable categories of AI risk in 2027, as DNS-layer blocking catches these attempts before an agent can act on them. That makes it one of the clearer wins security teams can point to as proof that AI risk is manageable, not just theoretical.

Frank Ruge, SVP EMEA Direct Sales

2025 was the year of experimentation. 2026 was the year of productivity. And 2027?

AI is being adopted faster than the cloud, mobile or the internet itself, and most enterprise plans are still drawn for a slower curve. In 2027, AI stops being a tool employees use and becomes a workforce that operates alongside them. Many large enterprises will have more software agents working in their systems than people, and boards will start asking a question few CEOs can answer today: how many agents do we employ, what can they touch, and who do they report to?
The AI projects that stall in 2027 won’t fail because of the models. They’ll fail on the plumbing: companies spending heavily on GPUs and platforms while having no clear view of what is resolving, connecting and talking to what. The first board-level AI post-mortem will blame something deeply unglamorous, such as a DNS misconfiguration, an address conflict or a flaky link between clouds. The winners will be the ones who knew exactly where every street was before they sent traffic down it.
Every AI dollar has to come from somewhere, and CIOs will find it by cutting tools. Point products that need a human to click through a console will be treated as legacy, and platforms that agents can drive through open APIs will take the budget. Expect procurement to add a new question to every renewal: can an agent operate this without us? Vendors who answer “not really” will lose the renewal long before they lose the technical evaluation.
Enterprises will not flip a switch to full autonomy, but they will climb quickly. Expect autonomy levels, much like those used for self-driving cars, to start appearing in RFPs, with customers asking vendors which capabilities run unattended, which need approval and which stay manual. The surprise is that the human-in-the-loop will become the bottleneck, and the companies that invest early in guardrails, audit trails and rollback will be the ones allowed to remove it first.
When launching a convincing, personalized attack costs close to nothing, a human-speed SOC cannot win by working harder. By the end of 2027, a security operation that can’t contain routine incidents without a human touch will look the way an unpatched server looks today: negligent. Expect the most forward-looking CISOs to report to their boards on the share of incidents contained automatically, and to be judged on it.
Regulators and customers in Europe will not accept “we think it’s fine” as an answer about AI. Between the EU AI Act, NIS2, DORA and growing sovereignty demands, enterprises will need a live inventory of their agents and models, where they run and what they do, with every action logged and reviewable. In tenders, “show me your agents” will sit next to “show me your certifications”. Companies that can answer in minutes rather than weeks will close deals that others can’t even bid on.
IT teams won’t grow at the pace of infrastructure, data and AI workloads, so the leverage has to come from somewhere else. The best IT teams of 2027 will look small on the org chart and operate far above their weight, because every engineer supervises a fleet of agents across network, security and cloud. The new measure of a strong IT organization won’t be headcount or tickets closed but how much infrastructure each person can safely run.