How AI is changing cybersecurity threats (and how it isn’t)

Getty Images

With AI dominating cybersecurity (and most other) headlines, it feels like it has rewritten the threat landscape. Deepfakes, prompt injection, attacks against large language models and autonomous, AI-powered hacks suggest we may be living through a major cyber inflection point

Some of these threats are real, particularly for organizations heavily utilizing AI agents. But the narrative often overshadows a more important reality: The biggest threats most businesses face aren’t from threat actors breaching defenses using new AI-based attacks. From what I’ve seen in global cyber incident response, successful attacks still rely on familiar, tried-and-true tactics

The most effective tools in threat actors’ toolkits remain phishing, credential theft, impersonation and social engineering. The difference is that AI helps execute these attacks faster, at scale and with a level of polish that was previously harder to achieve

Understanding the difference between overhyped AI risks andthreats security teamsface every day is critical for a practical approach to cybersecurity

Separating reality from AI hype

Despite fears that AI would eliminate technical barriers to cybercrime, successful attacks still require meaningful expertise. Take, for example, heavily discussedprompt injection attacks

These attacks involve manipulating an AI system to reveal information or using malicious prompts to cause it to perform actions it wasn’t intended to. While security researchers have demonstrated successful attacks — and organizations deploying AI tools should understand and guard against such risks — such an attack is not as simple or effortless as it may seem

Manipulating the enterprise AI systems most organizations use takes time, access to the target environment and a level of technical sophistication many threat actors don’t possess. Organizations using mature frontier AI models benefit from the substantial, built-in security controls designed to limit abuse

Recent events, such as theJuly 2026 security incidentinvolving OpenAI’s evaluation models and Hugging Face, demonstrate that advanced AI-enabled attacks aren’t purely theoretical. Today, however, they remain the exception

Organizations should address prompt injection without diverting attention from the identity and access security weaknesses behind most successful incidents

The same threats, supercharged

Emerging AI attack techniques, such as prompt injection, capture attention because they’re new and alarming. By comparison, ransomware feels like yesterday’s news. Yet, despite more than 15 years of headlines and defenses against it, ransomwareremainsone of the most disruptive and costly cyber risks organizations face

SMBs remain frequent ransomware targets because their data backup regimens are often less frequent or less thorough. While some threat actors may boast about the use of AI — often as a psychological tactic — the core methods behind effective ransomware attacks are familiar: phishing, stolen credentials or social engineering, credential harvesting, lateral movement and malware execution

Ransomware may be considered old news, but attacks still wreak havoc on organizations

Similarly, business email compromise is still common. Accessible tools like phishing kits mean even novice attackers can launch convincing phishing emails that direct recipients to fake login pages designed to steal credentials. Once inside, they can monitor activity, waiting for an opportune moment to intercept and profit

AI has proven to be anextremely valuable toolfor accelerating, scaling and making traditional attacks dramatically more convincing

If there’s one area where AI is demonstrably changing the threat landscape, it’s social engineering. The human attack surface remains one of the most vulnerable areas of an organization’s security posture

AI can create nearly undetectable phishing emails and fraudulent websites. AI agents can generate realistic fake documents and tailored communications with minimal effort, enabling threat actors to quickly adapt messages for specific industries or targets. If attackers have already compromised the email system, they’ll know when to send a perfectly timed but fraudulent payment request

Deepfake technology has further expanded these capabilities as communications that look and sound authenticconvincingly imitateexecutives, employees and business partners

Organizations should evolve their awareness programs beyond traditional red flags of suspicious language or mismatched sender domains. They need to follow established approval procedures, especially for financial transactions

AI’s threat doesn’t end with the attack

Another emerging challenge is the role AI playsafteran attacker gains legitimate network access, often through stolen credentials. Increasingly, enterprise AI agents give attackers a potential pathway to knowledge systems that aggregate information across the organization

Known as “living-off-the-agent,” threat actors can interact with agents of a corporate AI platform as an authorized user, asking questions and locating sensitive data. Such attacks are especially challenging because the activity often appears indistinguishable from legitimate user behavior. The attacker isn’t exploiting a vulnerability in the AI platform; they’re leveraging access already available.

Risks from employee use of AI

Lastly, unauthorized AI adoption by employees is particularly concerning. This year, 63% of2,000 respondentsto BlackFog’s shadow AI survey reported they believe it’s acceptable to use AI tools without employer approval if there is no company-approved option, with 58% using less secure free versions

The challenge is that organizations can’t govern AI tools they don’t know are in use

When employees input sensitive internal information or business data into unsanctionedAI platforms, they inadvertently create new points of exposure. With such widespread use, shadow AI presents a more immediate and practical risk than any complex AI exploit

Clearly defining employee AI use policies and maintaining visibility across the organization must be part of any AI governance strategy

Focus on what matters most

AI is transforming the cybersecurity landscape. But for most organizations, the greatest risks are not those capturing executive imaginations or industry headlines

Most successful incidents still begin with compromised identities, phishing campaigns, weak access controls and human error

Organizations can still achieve resilience by fortifying identity security, implementing robust multifactor authentication, improving phishing resistance and mindfully approaching AI deployments. The future of cybersecurity will involve AI, but addressing risks that attackers exploit today remains the foundation of a sound security strategy

About the Author

Source: www.informationweek.com