How AI Is Changing Identity Security: Redefining Credentials, Privileges, and Trust

What the Verizon 2026 DBIR Reveals About AI and Identity Security

For decades, cybersecurity teams have protected networks, endpoints, applications, and data. But attackers discovered something important:

The easiest way into an organization is not always through a vulnerability - sometimes it is through stolen credentials or a compromised trusted identity.

Now artificial intelligence (AI) is accelerating that reality. AI is changing both sides of cybersecurity. Defenders are using AI capabilities to improve detection, automate responses, apply machine learning, and analyze threats faster than ever before.

Attackers are using AI to:

• Scale reconnaissance
• Improve phishing
• Automate vulnerability discovery
• Create malware variants
• Impersonate trusted users
• Accelerate privilege escalation

The cybersecurity race has entered a new phase of AI versus AI. Identity determines who wins the battle.

The 2026 Verizon Data Breach Investigations Report (DBIR) analyzed more than 31,000 security incidents and over 22,000 confirmed breaches — the largest dataset Verizon has examined in a single report — providing one of the clearest pictures yet of how cybercrime continues to evolve.

Every identity is now part of the attack surface: employees, administrators, developers, applications, service accounts, workloads, APIs, and increasingly AI agents. 

The challenge is no longer simply managing traditional identities through identity and access management (IAM) systems. It is understanding identity risk across humans, machines, and AI agents.

AI Has Changed the Speed of Attack — Not the Goal

Attack techniques continue to evolve, but attacker objectives remain surprisingly consistent.

They want:

• Access
• Credentials
• Privileges
• Persistence
• Data

AI does not replace traditional cybercrime. It amplifies it by creating an AI-enhanced threat landscape where attackers operate faster, automate decisions, and adapt their techniques. 

A phishing email that previously took hours to research and write can now be generated and localized into the targeted victim’s language in seconds. Language itself is no longer a protection from phishing attacks. A social engineering campaign that required manual preparation can now be personalized automatically. A vulnerability that required deep expertise can now be analyzed with AI assistance.

AI Is Scaling Known Attack Techniques

The data shows how quickly the threat landscape is changing.

Exploitation of vulnerabilities became one of the dominant initial access vectors, climbing to 31% of the dataset - up from 20% the prior year, a 55% jump in a single reporting period. 

At the same time, only 26% of critical, known-exploited vulnerabilities were fully remediated by organizations in 2025, down from 38% the year before, and the median time to fully resolve them stretched to 43 days.

The DBIR confirms that threat actors are already using Generative AI across multiple stages of an attack, including:

• Target selection
• Gaining initial access
• Vulnerability research
• Malware development
• Tool creation

Analysis of AI-assisted threat activity shows attackers experimenting with AI models across multiple MITRE ATT&CK techniques, using them as a force multiplier for research, automation, scripting, and operational efficiency. 

Importantly, the DBIR’s own analysis found that AI’s primary impact today is operational, not novel: it is automating and scaling techniques defenders already know how to detect, rather than unlocking entirely new attack surfaces. Most AI-assisted malware and tooling mapped to well-known techniques, with a median of 55 existing malware examples performing the same function. 

Fewer than 2.5% of observations involved rare or novel techniques. 

AI-Powered Identity Attacks Are Changing the Attack Chain

Traditional attacks asked: “Where is the vulnerable system?”

Modern AI-driven attacks ask: “Who has access, and how can I become one of them?”

Step 1: AI Reconnaissance

AI-assisted tools can collect intelligence, perform user pattern analysis, and automate the discovery of:

• Employee information
• Technologies used
• Suppliers
• Executives
• Developers
• Exposed systems

Public information becomes attack intelligence.

Step 2: AI Social Engineering

Generic phishing is giving way to more personalized attacks.

AI enables:

• Personalized emails
• Realistic conversations
• Voice cloning
• Deepfake impersonation
• Multilingual attacks

The human element remains central to breaches, appearing in 62% of them this year, up slightly from 60% the year before.

Social engineering is the third most common breach pattern, present in 16% of all breaches. Phishing held steady at 16% as an initial access vector, while pretexting — the more involved, conversational form of social engineering — reached 6% and is increasingly used as the opening move in high-profile ransomware and extortion attacks.

In phishing simulations, non-email social engineering channels outperformed email. The median response rate for vectors such as voice calls and text messages was about 40% higher than for equivalent email campaigns. Large organizations also faced a median of 48 SMS-based phishing attempts against mobile devices per year.

Attackers no longer need perfect malware. They need a believable conversation built around realistic user behavior and trusted interactions.

Step 3: Credential Theft

Once trust is established, attackers target:

• Passwords
• MFA codes
• Session tokens
• API secrets
• OAuth permissions
• Authentication cookies

Here, the DBIR data tells a more layered story than “credential abuse is declining.” As an initial access vector, specifically, credential abuse fell from 22% to 13% this year. That drop is largely explained by the DBIR splitting out pretexting as its own tracked vector for the first time. Without that change, the comparable figure would be closer to 16%. 

Looked at differently, credential abuse still sits on top at 39% when every instance in the breach progression is counted, not just the opening move. 

Credentials also showed up as a stolen data type in 28% of breaches, and the use of stolen credentials remains the single most common attacker action overall, present in 36% of breaches.

The future of credential theft is not just stealing passwords. It is stealing authenticated trust.

Step 4: Privilege Discovery

Once inside, AI can help attackers answer:

Who am I?What access do I have?Where can I go next?

Exactly the same questions defenders should already be asking.

Attackers map:

• Group memberships
• Cloud roles
• Administrative privileges
• Identity relationships
• Service accounts

The DBIR’s own attack-graph data shows why this matters: 16% of organizations analyzed had roughly 80% exposure, meaning that once an attacker gained any low-level foothold, they had an 80% or better chance of reaching a key administrative account or piece of infrastructure.

Identity becomes both the attacker’s map and the defender’s opportunity.

The Rise of Autonomous Attackers

The next major shift is not simply AI-generated phishing. It is the movement toward more autonomous cyber operations.

AI agents will increasingly be able to:

• Identify targets
• Find exposed identities
• Test credentials
• Discover privileges
• Select attack paths
• Adapt techniques

AI allows attackers to operate at a scale and speed that were not previously possible. A human attacker might test hundreds of paths. An AI system can test thousands.

AI Agents Create a New Identity Security Problem

Organizations are rapidly adopting AI agents.

These systems will:

• Access applications
• Query databases
• Execute workflows
• Communicate with users
• Make decisions

But every AI agent needs an identity and permissions. Every permission must be understood and controlled. The next identity explosion will not come from employees.

It will come from AI systems that behave like digital insiders — operating with access, permissions, and decision-making capability.

Agentic AI and Machine Identities: The New Privileged Users

Today, organizations already struggle with:

• Service accounts
• API keys
• Certificates
• Automation scripts
• Cloud workloads

AI agents add another layer.

The DBIR’s own third-party cloud research underscores this point: looking at IaaS environments, 37% of organizations had an admin account with MFA disabled, and the report explicitly calls out that organizations should pay special attention to service and machine accounts, since those are the identities most likely to be leveraged in an agentic AI future.

The identity question evolves:

Not only “Who has access?”

But “What has access, what can it do, and should it still be trusted?”

Shadow AI is the New Shadow IT

The DBIR highlights another emerging challenge: employees adopting AI tools faster than organizations can secure them.

Key Data Points

• 45% of employees are now regular users of AI on corporate devices, up sharply from 15% the previous year
• 67% of users accessing AI platforms from corporate devices are doing so through non-corporate accounts, which puts that usage outside the organization’s visibility and control
• Shadow AI is now the third most common non-malicious insider action detected in DLP datasets - a fourfold increase over the prior year
• Source code was the most common data type submitted to external AI services, followed by images and other structured data
• In 3.2% of DLP policy violations, research and technical documentation were uploaded to unauthorized AI tools, raising real intellectual property exposure
• The average organization had more than 15% of users running unauthorized AI browser extensions — tools that can quietly collect and retain what employees browse, including internal sites

Shadow AI creates new risks:

• Sensitive data leakage
• Intellectual property exposure
• Unmanaged identities
• Uncontrolled integrations

How AI Accelerates Privilege Escalation

Attackers do not need full compromise immediately. They need a starting point and a normal identity, such as a forgotten account or a weak permission. Once compromised, attackers climb through the environment by expanding privileges.

The DBIR frames privilege escalation around four practical pillars:

1. Passwords
2. Configurations
3. Permissions
4. Patches. 

It maps these against MITRE ATT&CK’s Privilege Escalation and Credential Access techniques.

Mapping mitigations across those techniques, the report found that privilege management addresses roughly 65% of them, making it the single most impactful defense in this category, ahead of configuration hardening (33%) and password policy controls (30%). 

Patching alone mitigates only about 10% of these specific techniques, a reminder that vulnerability management matters, but it isn’t the main lever for stopping privilege escalation once an attacker already has a foothold.

This matters even more with AI because it can accelerate:

• Permission discovery
• Attack path mapping
• Privilege chaining
• Misconfiguration discovery

Key Data Points

• Credential dumping from memory (OS Credential Dumping / LSASS) was one of the most common techniques observed, appearing in 34% of threat intelligence data and 20% of incidents
• Users are more than four times more likely to be using an already-compromised password than what would be classified as a merely “weak” one
• Across attack-path data, 16% of organizations had roughly 80% exposure - meaning a low-level foothold had an 80%+ chance of reaching a key admin account
• 26% of organizations still had unresolved privilege escalation vulnerabilities dating back to 2021, and 11% had some from 2018

Why AI Makes Least Privilege Non-Negotiable

Traditional access model: “Give access because someone might need it.”

AI-era access model: “Prove access is required right now.”

Organizations must move toward:

• Just Enough Access - Only the permissions required.
• Just-in-Time Access - Only when required.
• Zero Standing Privilege - No identity should permanently hold powerful access without a business requirement.
• Continuous Verification - Trust must be continuously reassessed.
• Identity Intelligence - Understand relationships and risks.

How to Build Identity Resilience for the AI Era

The future security strategy must include:

1. Discover Every Identity

Including:

• Humans
• Admins
• Contractors
• Service accounts
• APIs
• AI agents

Unknown identities become opportunities for attackers.

2. Build an Identity Graph

Understand:

• Who has access
• Why they have access
• Where privileges connect
• What paths attackers could abuse

Attackers think in paths, so defenders need to see paths.

3. Eliminate Standing Privilege

4. Secure AI Agents Before They Scale

Every AI agent needs:

• Ownership
• Authentication
• Authorization
• Monitoring
• Lifecycle management

Never deploy an AI agent without a known identity and owner.

5. Treat Credentials Like Vulnerabilities

A compromised identity should trigger the same urgency as a critical CVE.

Ask:

• Where was it used?
• What did it access?
• What privileges existed?
• What connected systems trusted it?

6. Detect Identity Threats Continuously

Organizations must detect:

• Impossible travel
• Privilege changes
• Abnormal access
• Token abuse
• Suspicious automation
• Unusual AI agent behavior

Identity security must move from periodic reviews to continuous intelligence using behavioral analytics, anomaly detection, and adaptive identity controls.

The DBIR's infostealer research reinforces why speed matters here: 73% of ransomware victims had a prior infostealer infection or credential leak in the year before the attack, and for half of those, the credential event happened within just 95 days of the ransomware hitting.  

A credential leak is rarely the end of the story - treat it as an early warning.

The Future: AI Will Attack at Machine Speed and Identity Must Defend at Machine Speed

The future of cybersecurity will not simply be humans versus hackers.

It will be AI attackers versus AI defenders. Autonomous systems competing for advantage.

But the winner will be determined by control of identity.

Because every action requires an identity. Every identity carries privilege. Every privilege creates risk. 

The organizations that succeed in the AI era will be those that know:

• Every human identity
• Every machine identity
• Every AI agent
• Every privilege
• Every trust relationship

AI changes the speed of cybersecurity, and identity determines the outcome.

For years, we said identity was the new perimeter. In the AI era, identity becomes something bigger.

Identity is the control plane for digital trust. Every human, machine, and AI agent must be known, governed, monitored, and continuously verified.

Put Privileged Access Under Control Before AI Speeds Up the Risk

AI is making it faster to find credentials, discover privileges, and move through connected systems. Security teams need to know who and what has privileged access, limit unnecessary permissions, and see when that access is being used.

Segura® PAM helps organizations control privileged access, secure credentials, reduce standing privilege, monitor activity, and create a clear record of who accessed what.

Explore Segura® PAM to see how your team can strengthen identity security as attacks move faster.

*Statistics and findings referenced throughout are drawn from the Verizon 2026 Data Breach Investigations Report.