5 predictions that will define security in 2027

A pragmatic guide to managing voice phishing, containment speed, and rogue AI agents.

À propos de l’auteur

Matt Immler

Regional CSO

19 novembre 2026 Temps de lecture: ~

Everyone is bringing AI into their organization right now, and security budgets are following close behind. At the same time, many organizations assume their security foundations are already covered because they’ve rolled out strong controls at the front door. 

That’s where companies are going to get caught off guard in the new year. 

I don’t think we’re going to see a massive technological leap with AI in 2027, which means the biggest risks won’t come from some brand-new AI attack vector. They’ll most likely come from the identity and access gaps we already have: the MFA exception paths, the help desk workarounds, and the places where users, tokens, and AI agents still have too much standing access.

Based on my daily conversations with customers, here are my security predictions for 2027 and how companies can prepare for them.

1. One-and-done security strategies will lead to big regrets

In 2027, organizations that rely on a small number of security controls will probably wish they had done more, and sooner. Enrolling an entire company in phishing-resistant multi-factor authentication (MFA), such as passkeys, is a good start. But that won’t be enough if other weak links remain. 

Attackers look for the path of least resistance. Even when organizations enroll their entire workforce in strong, phishing-resistant authentication like passkeys, attackers can target the exception paths to execute an MFA downgrade attack. Weak fallback options, such as one-time codes and social engineering at the help desk, give threat actors a way around your front-door defenses. And with AI in the mix, finding and exploring those paths has only gotten easier. Building resilience in the coming year means engineering and testing exception paths, such as lost devices and help desk enrollment, with the same rigor as the front-door sign-in path itself.

2. Fake IT support calls will put employees on the spot 

While security teams are accustomed to defending against credential stuffing, brute-force attacks, and traditional email phishing, real-time voice phishing (vishing) continues to catch organizations off guard.

The human element is involved in 62% of security breaches, according to Verizon’s 2025 Data Breach Investigations Report, and attackers eagerly exploit this vulnerability through high-pressure voice calls. When an attacker calls an employee pretending to be an IT support technician, they put the victim on the spot, creating a false sense of urgency that bypasses traditional anti-phishing training. 

One way companies can fight back is by combining employee awareness with the correct security controls, like limiting how new MFA factors are enrolled by restricting registrations to company-managed devices or trusted network zones.

3. Budgets will shift toward security for AI—and AI for security

AI spending is a top priority for security leaders: 69% of CISOs are prioritizing AI-related security in the next budget cycle, according to the 2026 IANS and Artico Search Security Budget Benchmark Report. 

Security budgets will follow the failures teams need to prevent—both in securing AI deployments and using AI in the SOC. For example, companies might fund AI-driven security alert triage to help analysts make better, faster decisions. At Okta, our security team does this for low-level alerts by using automated or AI-type prompts to ask users, "Was this you?" If the user confirms the action, the alert clears. This straightforward approach keeps teams from wasting hours on low-level alerts, freeing up valuable resources to focus on actual complex threats.

4. Containment will define risk for both human and AI identities

Organizations rightly focus on stopping attackers at the front door. But identity also shapes what happens after someone gets in. When a bad actor takes over a user session, the damage depends on what that identity can reach. Can a stolen credential reach an admin console? Does a former contractor still have access to a SaaS application? Can a vendor token pull every customer record?

Because of this exposure, security success in 2027 will be defined by containment speed, not just a secure perimeter. The priority must shift to how quickly a team can cut off a compromised identity, terminate active sessions, and restrict downstream access—particularly for autonomous agents operating at machine speed.

But if an agent goes rogue or is tricked into unauthorized actions, simply shutting down the system is a mistake because it destroys valuable forensic evidence. Instead, the trend will shift toward dedicated AI kill switches. These switches cut off the agent’s specific access and privileges, allowing security teams to investigate who authorized the action and how the compromise occurred.

5. AI agents will optimize their way to a breach

Because AI agents are non-deterministic, they will take any path to complete a task (emphasis on the word “any”). That’s why before the year is out we are bound to see a Fortune 500 company publicly disclose a breach caused by an autonomous AI agent tricked into leaking sensitive customer data.

Without the right identity guardrails, an agent tasked with formatting customer data might export an entire database to an insecure third-party website just to get the job done. To prevent these unexpected detours, organizations are best served by establishing clear boundaries on what data an agent can access and where it is allowed to send it. Aligning with frameworks like the Blueprint Alliance reference architecture—an open, multi-vendor framework that shifts security from the network perimeter to a unified, Zero Trust, agentic control plane—gives technology leaders a concrete way to scale agentic deployments without losing control.

This content is for informational purposes only, does not constitute professional advice, and is provided without warranties or liability; please consult your own advisors for implementation decisions. 

 

À propos de l’auteur

Matt Immler

Regional CSO

Découvrez notre newsletter sur l'identité

Newsletter Access Granted