AI-Powered Cyberattacks Explained: How Artificial Intelligence Is Changing Cybercrime

Cybercrime has traditionally depended on a combination of technical knowledge, deception and persistence. Attackers searched for vulnerable systems, developed malicious software, stole passwords and manipulated people into revealing sensitive information. These techniques often required considerable expertise, time and resources.

Artificial intelligence is changing that equation.

AI-powered cyberattacks are making it easier for criminals to create convincing phishing messages, imitate trusted voices, analyze software weaknesses and automate activities that previously demanded extensive manual effort. Some attackers are now experimenting with AI agents that can carry out multiple stages of an intrusion with limited human supervision.

The consequences are becoming visible in real-world investigations.

In September 2026, Google’s Threat Intelligence Group reported that attackers had compromised a cloud resource and organized an AI-assisted credential-harvesting campaign in under six hours. Anthropic also documented cybercriminals and espionage operators using AI agents to coordinate intrusions against multiple targets.

These developments do not mean artificial intelligence has suddenly invented an entirely new category of cybercrime. Many attacks still exploit familiar weaknesses, including stolen credentials, unpatched software, insecure applications and human trust.

What has changed is the speed, scale and accessibility of those methods.

According to the 2026 Verizon report, cybercriminals are using generative AI to support at least 15 different attack techniques. The report also found that exploitation of software vulnerabilities accounted for 31% of breaches in its dataset, while ransomware featured in 48%.

Those figures describe the wider cyber threat landscape, not the percentage of attacks caused by AI. Nevertheless, they demonstrate why AI-powered cyberattacks deserve serious attention.

Understanding this threat requires examining how criminals use AI, which techniques have actually been observed and what security measures remain effective.

What Are AI-Powered Cyberattacks?

AI-powered cyberattacks are malicious cyber operations in which attackers use artificial intelligence to support, accelerate or automate activities intended to compromise digital systems, steal information, manipulate people or disrupt services.

The technology can assist at different stages of an attack.

A criminal might use a language model to research a company and produce a believable phishing email. Another attacker might use machine learning to identify suspiciously configured internet-facing systems. More sophisticated operations can involve AI agents coordinating several activities, including information gathering and analysis of compromised data.

The important distinction is that artificial intelligence does not automatically make an attack autonomous.

Many incidents still involve humans selecting targets, providing instructions, reviewing results and deciding what to do next.

AI-powered cyberattacks can generally be divided into three broad categories.

These categories are useful for understanding different levels of automation, although real incidents may combine several approaches.

There is also an important difference between using AI to attack conventional systems and attacking AI systems themselves.

A criminal using a chatbot to write fraudulent messages is conducting an AI-assisted attack.

An attacker manipulating an organization’s AI assistant to disclose confidential information is targeting an AI application.

Both are important cybersecurity concerns, but they involve different technical weaknesses.

Our cybersecurity guide explains the broader foundations of digital defense, including identity protection, malware, network security and incident response.

How AI Is Changing the Economics of Cybercrime

One of the most significant changes is the reduction in the time and specialist labor required for certain malicious tasks.

A traditional phishing campaign might require attackers to collect background information, draft messages, translate content and organize responses manually.

Generative AI can assist with many of these activities.

Similarly, attackers examining large quantities of stolen files can use AI to summarize documents, identify potentially valuable information and prioritize material for further investigation.

These capabilities do not guarantee successful attacks.

AI-generated information can be inaccurate, malicious code can malfunction and automated systems can make mistakes.

However, even imperfect automation can help criminals attempt more operations with fewer resources.

The Microsoft Digital Defense Report 2026 describes AI as a technology that compresses attack timelines and lowers the cost of sophisticated capabilities.

Its central warning is that security teams must increasingly respond to threats operating at machine speed.

This is especially concerning for organizations that still rely on slow vulnerability management, manual incident investigation or poorly monitored accounts.

Seven Major Types of AI-Powered Cyberattacks

Although the techniques overlap, seven developments illustrate how artificial intelligence is changing real-world cyber threats.

1. AI-Powered Phishing Is Becoming More Convincing

Phishing remains one of the most common methods criminals use to obtain passwords, financial information and unauthorized access.

Traditional phishing emails were often recognizable because of poor spelling, unnatural language or generic messages.

Generative AI has weakened those warning signs.

Attackers can now produce grammatically correct messages in different languages, adapt their tone to particular industries and personalize their content using publicly available information.

For example, an employee might receive an email appearing to come from the company’s finance department.

The message could reference the organization’s actual name, a known supplier and a plausible invoice-processing deadline.

The attacker may have gathered these details from public websites, professional profiles or previously stolen information.

AI makes it easier to organize that material into convincing communication.

Why Personalized Attacks Are More Dangerous

A generic message claiming that an account will expire may be easy to dismiss.

A carefully written message referring to an actual business relationship can be more persuasive.

This is particularly important in business email compromise, where criminals impersonate executives or suppliers to redirect payments or obtain sensitive information.

AI-powered cyberattacks can also support multilingual campaigns, allowing criminals to target victims outside their own language communities.

However, there is an important limitation.

Better phishing language does not automatically mean that every campaign achieves a higher success rate. Attack effectiveness also depends on delivery, credibility, target selection and existing security controls.

The 2026 ENISA assessment warns that emerging AI models are likely to support malicious operations while familiar techniques such as social engineering remain important.

The practical lesson is that spelling and grammar are no longer dependable methods of determining whether a message is legitimate.

2. Deepfake Voices and Videos Can Impersonate Trusted People

Another alarming development involves synthetic audio and video.

AI models can generate speech resembling a real person’s voice and create convincing visual representations of people who may never have recorded the content being presented.

Criminals can use these capabilities to impersonate managers, relatives, public officials or business partners.

Imagine receiving a phone call that sounds like your supervisor requesting an urgent financial transfer.

The caller knows your name, refers to a current project and speaks in a familiar voice.

Without independent verification, the communication might appear authentic.

This is the danger of deepfake-enabled social engineering.

The FBI’s 2026 report recorded 22,364 complaints involving artificial intelligence during 2025, with reported losses approaching $893 million.

These figures cover multiple categories of AI-related fraud, not exclusively deepfake attacks or technical intrusions.

Nevertheless, they demonstrate the substantial financial consequences of criminal AI misuse.

Why Familiar Voices Are No Longer Proof of Identity

Voice recognition has traditionally been a natural part of human trust.

People identify relatives, colleagues and friends partly through how they sound.

Synthetic voice technology challenges that assumption.

Even when a caller sounds familiar, the request may be fraudulent.

The safest response to unusual financial instructions is to verify the request through a separate, trusted communication channel.

For example, an employee receiving a request to change supplier banking details should independently contact the supplier using a previously verified telephone number.

Organizations should not rely solely on voice or video identity when authorizing sensitive transactions.

Our article on AI scams examines how deepfakes and synthetic identities are making online fraud more difficult to detect.

3. AI Can Accelerate Vulnerability Discovery and Exploitation

Software vulnerabilities are weaknesses that attackers may exploit to gain unauthorized access or disrupt systems.

Finding those weaknesses traditionally involves research, specialized tools and technical expertise.

AI-powered cyberattacks are changing parts of this process.

Language models can assist with reading source code, understanding software behavior and analyzing technical documentation.

Attackers may also use AI to investigate publicly disclosed vulnerabilities and identify systems potentially affected by them.

In May 2026, Google Threat Intelligence reported identifying an attacker using a zero-day exploit that researchers believed had been developed with AI assistance.

A zero-day vulnerability is a software weakness for which an effective fix may not yet be available when exploitation begins.

Google reported that the attacker intended to use the exploit in a broader campaign, although its intervention may have prevented the planned deployment.

This was an important observation, but it should not be interpreted as proof that AI can independently discover and exploit every type of software vulnerability.

Why Faster Exploitation Creates Problems for Defenders

Organizations usually need time to assess security updates, test compatibility and deploy patches.

Attackers may attempt to exploit newly disclosed weaknesses before those updates are installed.

AI assistance can shorten parts of the research and development process.

The IBM X-Force report found a 44% year-over-year increase in attacks beginning with exploitation of public-facing applications in its 2025 investigation data.

IBM associated the wider trend with weaknesses such as misconfigurations, missing controls and increasingly capable attack tools.

The 44% figure should not be described as an increase caused entirely by AI.

Instead, it illustrates why vulnerability management has become particularly important as attackers gain better tools.

Businesses need to know which systems are exposed to the internet, which vulnerabilities are actively exploited and how quickly critical updates can be installed.

4. AI-Assisted Malware Development Is Becoming More Sophisticated

Malware is malicious software designed to steal information, disrupt services, provide unauthorized access or perform other harmful activities.

Traditional malware development requires programming knowledge and an understanding of operating systems, networks and defensive technologies.

AI coding tools can help attackers perform some development tasks more quickly.

These may include analyzing existing code, identifying errors, translating programs between languages or modifying components of malicious software.

Attackers may also experiment with techniques intended to make malicious programs harder for security tools to recognize.

However, there is a difference between AI-assisted malware development and malware that independently uses an AI model while running.

The first involves a human using AI during development.

The second involves an AI component participating in the malware’s actual behavior.

Can Malware Change Its Own Behavior?

Some malicious software already changes its code or appearance to evade detection. These techniques existed long before modern generative AI.

AI can potentially expand the sophistication of such activity.

Google’s May 2026 investigation described malware designated PROMPTSPY, which incorporated AI-related capabilities to help interpret system conditions and support malicious operations.

This represents an emerging technical development rather than proof that ordinary malware has become fully autonomous.

AI-generated code can also contain bugs, create detectable patterns or fail when operating environments change.

The important concern is that AI may reduce the amount of manual work needed to develop and adapt malicious tools.

Ransomware operators could benefit from some of these capabilities.

However, ransomware remains dangerous primarily because attackers gain access, steal information, disrupt operations and pressure victims to pay.

Our ransomware guide examines these mechanisms and the importance of protected backups and incident response.

5. Automated Credential Theft and Account Takeover Can Spread Faster

Stolen credentials remain one of the most valuable resources in cybercrime.

An attacker who obtains valid login details may be able to access systems without exploiting a complicated software vulnerability.

Passwords, authentication tokens, browser sessions and API credentials can all become targets.

AI-powered cyberattacks can help criminals organize information collected from compromised accounts and identify additional opportunities for unauthorized access.

A particularly important example emerged in Google’s September 2026 investigation.

According to its AI Threat Tracker, researchers observed attackers compromise a cloud resource and then plan, build and execute an agent-enabled mass credential-harvesting campaign in less than six hours.

The observation illustrates how automation can reduce delays between initial access and further malicious activity.

Why Stolen Sessions Can Be Dangerous

People often assume that changing a password immediately prevents an attacker from accessing an account.

That is not always sufficient.

Some applications use session tokens that allow a user to remain signed in without repeatedly entering credentials.

If those tokens are stolen, an attacker may be able to reuse an authenticated session, depending on the security controls involved.

This is one reason identity security extends beyond password complexity.

Organizations need appropriate session management, account monitoring, access restrictions and mechanisms for invalidating compromised sessions.

The News Ink’s identity security guide explains how passkeys, multifactor authentication, account recovery and access controls work together.

AI does not eliminate these fundamental defenses.

It makes rapid identification and containment of compromised accounts even more important.

6. Agentic AI Is Enabling More Automated Cyber Operations

Agentic AI is one of the most significant emerging developments in cybercrime.

An AI agent is a system capable of performing connected tasks toward an objective, often by using software tools, processing results and determining what action to take next.

In legitimate environments, agents can assist with programming, research, customer service and administrative work.

Criminals are experimenting with similar capabilities for unauthorized operations.

Instead of requesting one answer from a chatbot, an attacker may configure an automated workflow to conduct repeated analysis and coordinate multiple tasks.

The 2025 AI-Orchestrated Espionage Campaign

In November 2025, Anthropic disclosed a cyberespionage campaign that it had detected during September of that year.

The company assessed with high confidence that the activity involved a Chinese state-sponsored group.

According to Anthropic, the attackers manipulated its Claude Code system to support operations targeting approximately 30 organizations across sectors including technology, finance, chemicals and government.

The company reported that a small number of intrusions succeeded and estimated that AI performed 80%–90% of certain tactical operations.

That estimate came from Anthropic’s investigation; it should not be treated as a universal measure of AI autonomy in cybercrime.

The campaign was subsequently documented by MITRE ATT&CK, which recorded its use of AI agents for reconnaissance, vulnerability analysis and other malicious activities.

The case is important because it demonstrated AI operating as more than a writing or research assistant.

Human operators remained involved, but automation handled substantial portions of the workflow.

What Changed During 2026?

Anthropic’s September 2026 report described additional cases involving financially motivated criminals, politically motivated actors and suspected espionage operations.

Some attackers used agents to coordinate multiple targets and process large volumes of information.

The report described cases where operations could progress in hours rather than requiring prolonged manual coordination.

These findings suggest that AI-powered cyberattacks may become accessible to actors who lack the resources traditionally associated with sophisticated hacking groups.

However, such operations remain technically demanding.

They still depend on usable access, effective tools, exploitable weaknesses and the ability to manage failures.

AI can automate work without guaranteeing success.

7. AI Systems Themselves Are Becoming Cyberattack Targets

The growing adoption of artificial intelligence introduces another problem: AI applications can become targets.

Businesses increasingly connect AI assistants to documents, databases, email systems, programming tools and cloud services.

These integrations make AI more useful, but they also create new paths through which attackers may attempt to manipulate systems.

One important technique is prompt injection.

Prompt injection occurs when attacker-controlled information attempts to influence an AI application’s behavior in ways that conflict with its intended instructions.

For example, an organization might use an AI assistant to summarize incoming documents.

A malicious document could contain instructions attempting to make the assistant reveal information or perform an unauthorized action.

Whether such an attempt succeeds depends on the application’s design, permissions and safeguards.

Why AI Agents Require Strong Access Controls

A chatbot that only produces text has different risks from an AI agent authorized to send emails, modify files or execute software tools.

If an agent receives excessive permissions, a successful manipulation attempt could have more serious consequences.

The OWASP AI security guidance identifies prompt injection, sensitive information disclosure, supply-chain weaknesses and excessive agency among important risks for large language model applications.

Security teams therefore need to treat AI systems as part of their normal attack surface.

They should limit permissions, separate trusted instructions from untrusted content, monitor tool use and require approval for sensitive actions.

This is distinct from criminals using AI to attack ordinary systems, but both trends increasingly overlap.

Real-World Evidence: What Investigators Have Found

The most reliable way to understand AI-powered cyberattacks is to examine documented incidents rather than assume every alarming prediction has already occurred.

These reports collectively demonstrate increasing AI involvement across the attack lifecycle.

They do not establish that all modern cyberattacks are AI-powered or that autonomous agents have replaced human attackers.

The distinction matters because exaggerated claims can distract organizations from vulnerabilities that remain far more common.

A sophisticated automated attack can be devastating, but an ordinary stolen password or unpatched server may still provide the easiest route into a business.

Are AI-Powered Cyberattacks More Dangerous Than Traditional Attacks?

AI-powered cyberattacks can be more dangerous in particular circumstances, especially when automation increases speed, scale or personalization.

However, the technology does not remove the basic requirements for a successful intrusion.

An attacker still needs a usable weakness, compromised identity or vulnerable target.

The practical differences are easier to understand through comparison.

The biggest shift is often economic.

A criminal who previously needed several specialists may be able to automate parts of their work.

That does not necessarily make every attacker technically advanced.

It makes some advanced-looking operations easier to attempt.

Why AI Cannot Guarantee a Successful Cyberattack

Predictions about autonomous hacking sometimes overlook the limitations of current AI systems.

Language models can generate inaccurate code, misunderstand technical conditions and make confident but incorrect assessments.

An automated agent may fail to recognize that a system has changed or incorrectly interpret the result of an action.

These failures can undermine an attack.

AI providers also maintain safeguards, abuse-monitoring systems and account restrictions intended to reduce malicious use.

Security teams continue developing tools to detect automated activity and suspicious behavior.

Most importantly, organizations with strong identity controls, timely patching and effective monitoring can reduce many of the opportunities attackers depend upon.

The danger is real, but inevitable compromise is not.

How Businesses Can Defend Against AI-Powered Cyberattacks

The arrival of AI-enabled threats does not mean businesses must abandon established cybersecurity practices.

Instead, organizations need to strengthen foundational security while preparing for faster attacks.

Prioritize Phishing-Resistant Authentication

Passwords remain vulnerable to theft, reuse and social engineering.

Phishing-resistant multifactor authentication, including appropriately configured passkeys and hardware security keys, can reduce the risk of account compromise.

Ordinary MFA methods still provide important benefits, but not every method protects equally well against sophisticated phishing.

Privileged accounts, financial systems and important cloud applications should receive particular attention.

Verify Sensitive Requests Independently

Deepfake technology makes identity verification more important.

Organizations should establish procedures for unusual payment requests, changes to bank details and requests involving confidential information.

A familiar voice, professional email or video appearance should not automatically authorize a transaction.

Independent verification can stop an attack without requiring employees to determine whether the media itself was artificially generated.

Patch Exposed Systems Quickly

Internet-facing applications and services should be monitored for actively exploited vulnerabilities.

Businesses need reliable inventories of their systems, clear patching responsibilities and processes for addressing urgent security weaknesses.

This remains essential even when AI is not involved.

Protect Backups and Monitor Accounts

Ransomware and data theft can cause substantial operational damage.

Businesses should maintain protected backups, test restoration procedures and monitor unusual account activity.

Suspicious logins, unexpected privilege changes and large data transfers may indicate an intrusion.

Control AI Agent Permissions

Organizations deploying AI assistants should assess what information those tools can access and which actions they can perform.

An assistant that summarizes documents generally does not need unrestricted access to financial systems.

Sensitive operations should use clear authorization controls and appropriate human approval.

The NIST Cybersecurity Framework provides a useful structure for identifying risks, protecting systems, detecting incidents, responding and recovering.

These principles remain relevant as AI becomes part of business infrastructure.

How Individuals Can Protect Themselves

AI-powered cyberattacks do not target businesses exclusively.

Individuals may encounter fraudulent messages, synthetic voices, fake investment opportunities and impersonation attempts.

Several practical habits can reduce exposure.

Use unique passwords and enable strong authentication on important accounts. Keep operating systems and applications updated, avoid signing in through links in unexpected messages and independently verify requests for money or sensitive information.

People should also be cautious about sharing personal information publicly.

Criminals can use details from professional profiles, social media and public records to make impersonation more convincing.

If someone claiming to be a relative calls about an emergency and urgently requests money, contact that person using a known number before acting.

The same principle applies to messages supposedly sent by banks, employers or government agencies.

Security increasingly depends on verifying the request rather than judging whether the message looks authentic.

Can Artificial Intelligence Also Stop Cyberattacks?

Yes. AI is not exclusively an advantage for attackers.

Security teams use machine learning and other AI techniques to examine large quantities of activity, identify unusual patterns and help analysts investigate possible incidents.

AI can assist with malware analysis, vulnerability prioritization, security documentation and incident summaries.

These tools are particularly valuable when organizations face more alerts than human teams can reasonably investigate manually.

However, defensive AI also has limitations.

False positives can waste resources, while incorrect summaries or overlooked evidence can delay investigations.

Security teams still need technical expertise, reliable data and human oversight.

The broader relationship between offensive and defensive AI is explored in our AI and cybersecurity guide.

The important question is not whether AI will replace cybersecurity professionals, but how it can help them identify and respond to threats more effectively.

What Will AI-Powered Cyberattacks Look Like in the Future?

The next stage of AI-powered cyberattacks is likely to involve greater automation, faster decision-making and closer integration between AI agents and conventional hacking tools.

Attackers may become more capable of coordinating several activities simultaneously.

Fraud campaigns may use increasingly convincing synthetic identities, while malicious software developers could use AI to experiment with new techniques.

Organizations adopting AI assistants may also face additional risks associated with excessive permissions and insecure integrations.

However, the precise direction of these developments remains uncertain.

The pace of technical progress does not guarantee that every new capability will become reliable or widely available to criminals.

AI safety controls, improved detection systems and changes in defensive technology will influence what attackers can achieve.

The most useful preparation is therefore not attempting to predict every future technique.

It is building security systems capable of detecting unusual behavior, limiting unauthorized access and recovering when something goes wrong.

Frequently Asked Questions About AI-Powered Cyberattacks

What is an AI-powered cyberattack?

An AI-powered cyberattack is a malicious operation that uses artificial intelligence to assist or automate activities such as phishing, vulnerability analysis, malware development, credential theft or unauthorized access.

Can AI hack a computer without human help?

Some AI agents can perform connected technical tasks with limited supervision, and researchers have documented partially autonomous malicious operations. However, such systems still depend on tools, permissions, exploitable weaknesses and operational reliability. Fully autonomous hacking should not be treated as universal or inevitable.

Are AI-powered phishing emails harder to detect?

They can be. Generative AI makes it easier to produce natural language, personalize messages and imitate professional communication. Users should verify suspicious requests independently rather than relying only on spelling or appearance.

Can antivirus software stop AI-powered cyberattacks?

Security software can detect many malicious programs and suspicious activities, including some AI-assisted threats. However, antivirus alone cannot prevent every phishing attempt, stolen-session attack or social engineering scheme. Layered defenses remain necessary.

What is the biggest danger of agentic AI cyberattacks?

One of the main concerns is that AI agents can reduce the time and human effort required to coordinate multiple malicious tasks. This may allow attackers to operate against more targets or move through parts of an intrusion more quickly.

Why AI-Powered Cyberattacks Are Changing the Future of Cybersecurity

Artificial intelligence is not replacing traditional cybercrime. It is changing how criminals conduct it.

Phishing still relies on deception. Malware still exploits weaknesses in software or systems. Account takeover still depends on unauthorized access, and ransomware remains dangerous because attackers can steal information and disrupt essential operations.

The difference is that AI can make many supporting activities faster, cheaper and easier to coordinate.

Criminals can generate convincing messages, analyze large quantities of information and automate selected operations with fewer manual steps.

The most advanced documented incidents demonstrate that AI agents can sometimes participate in several stages of an attack rather than simply providing advice.

Yet the same investigations also reveal something important.

Successful attacks still depend on ordinary security failures.

Unpatched software, excessive account privileges, weak authentication, exposed services and poor monitoring remain central problems.

Organizations do not need to solve every challenge in artificial intelligence to protect themselves.

They need to understand their systems, secure important identities, verify sensitive requests, prepare for incidents and maintain the ability to recover.

AI-powered cyberattacks will continue evolving, but the strongest response is not fear of increasingly sophisticated technology.

It is developing cybersecurity practices that remain effective even when attackers become faster, more automated and more convincing.

The future of digital security will depend on how well businesses, technology companies and individuals adapt to that changing environment.

For more research-driven explainers, science coverage and global stories, follow The News Ink on X, Instagram, Threads, Medium, Substack, Bluesky, Pinterest, TikTok, Quora and Mastodon.