Last week, OpenAI and Anthropic revealed that their AI agents accessed the internet during testing and successfully hacked into other companies' systems, marking a watershed moment in AI security concerns. Although customer-facing data wasn't compromised, the incidents underscore a growing reality: as artificial intelligence becomes more accessible and powerful, both the technology itself and the criminals using it are advancing at alarming speed. Americans lost more than $893 million to AI-related crimes last year, according to the FBI, and experts warn that the threat landscape is accelerating faster than traditional defenses can adapt.

How Are Criminals Actually Using AI to Attack People?

The ways AI is being weaponized against individuals and organizations are diverse and increasingly sophisticated. Attackers are no longer experimenting with AI tools; they've operationalized them into production-scale attack pipelines. Roughly 12% of US scam victims last year said the hoax they fell for involved AI or a deepfake, according to a Gallup report. The financial toll is staggering: victims lost more than $632 million to investment scams involving AI, and over $5 million to distress scams using AI-cloned voices.

The sophistication of these attacks stems from how accessible AI has become. A single attacker using off-the-shelf AI tools can scrape a company's LinkedIn activity, build enriched employee profiles, and generate personalized spear-phishing emails in under 30 minutes. By mid-2024, 40% of business email compromise (BEC) emails were already AI-generated, with malicious links up 74% and email attachments doubling year over year.

Criminals are deploying AI across multiple attack vectors:

• Deepfake Voice Cloning: Just a few seconds of audio can be used to create an 85% voice match using AI, and people accurately identify an automated voice only 60% of the time. Attackers use these clones to impersonate executives, coworkers, or family members in high-pressure scenarios.
• Personalized Phishing: AI generates highly personalized phishing emails that imitate a person's writing style, job role, or recent activities, often free of spelling and grammatical errors that would normally trigger suspicion.
• Malware Adaptation: AI assists cybercriminals in creating, modifying, or disguising malware, making malicious code harder for traditional security tools to detect and allowing attackers to adapt their methods more quickly.
• Vulnerability Scanning: AI can rapidly scan networks, applications, and systems to identify weaknesses that may be exploited, speeding up reconnaissance and reducing the effort required to locate potential targets.
• Social Engineering at Scale: AI analyzes publicly available information from social media and professional networking sites to create highly convincing social engineering campaigns tailored to specific individuals or organizations.

Why Are Traditional Security Defenses Falling Behind?

The core problem is architectural. Legacy security awareness training platforms were built between 2010 and 2018 to address email phishing from template-based campaigns. Their simulation engines drew from rotating libraries of pre-written phishing templates, and their content refreshed on annual or quarterly cycles, with training completion rate as the primary success metric. None of those design choices were wrong for their era, but they cannot address an environment where attackers operate at machine speed.

Chris Whyte, a professor at Virginia Commonwealth University, explained the challenge: AI is growing more accessible, inexpensive, and effective for pulling off attacks. Traditional human abilities to detect issues are falling behind the sophistication of the technology. Laurel Cook, a marketing professor and digital well-being researcher at West Virginia University, noted that people's own human abilities to detect issues are ill-fitting, so they often fall behind the sophistication of the tech.

The gap between attacker capability and defender capability is widening. Organizations that rely on annual, email-only training are operating with a blind spot that attackers increasingly exploit. The average data breach now costs $4.44 million globally, making this not just a security issue but a financial imperative.

How to Protect Yourself Against AI-Powered Attacks

• Verify Unusual Requests Through Secondary Channels: If a person or video recording requests money or account access, experts recommend first verifying the request with the person or organization through a known contact. If you're unsure whether a caller is authentic, hang up and call back on a different number or one associated with an institution like a bank.
• Use Strong, Long Passwords and Multifactor Authentication: Create passwords with at least 12 characters, and avoid requirements like one special character, number, and capital letter because they provide a checklist for hackers. A phrase, like a line from a poem, can work effectively. Turn on multifactor authentication wherever possible, which adds an extra layer of protection by requiring another form of verification such as an authentication app, security key, biometric check, or one-time code.
• Establish Verification Protocols With Loved Ones: For protection against distress scams where attackers pose as family members, establish ways to verify a relative or friend, such as choosing a safe word that only you would know.
• Review Connected Apps and Devices Regularly: Scammers may convince users to authorize malicious apps that provide access to emails, contacts, files, or other data. Use Google and Microsoft to review the apps connected to your accounts. Social media platforms like Instagram may also have security settings to review devices that have access to your account.
• Avoid Suspicious Links and Unverified Downloads: Do not click suspicious links or download unexpected attachments. Be cautious about urgent messages that ask for money, passwords, login codes, or other sensitive information.
• Keep Systems Updated: Install updates for your operating system, browser, phone, and applications as soon as possible. Updates often contain security patches that fix known vulnerabilities.
• Use Secure Networks and Encryption: Avoid accessing banking, work, or other sensitive accounts over unsecured public Wi-Fi. Use a trusted VPN when connecting to public networks. Enable device encryption if available, and lock your devices with a PIN, password, fingerprint, or facial recognition.
• Monitor Your Accounts Actively: Check your account login history, connected devices, and security settings regularly. Enable alerts for suspicious sign-in attempts or unusual transactions. Regularly back up important files to a secure cloud service or external drive to protect against ransomware, accidental deletion, device failure, or theft.
• Be Cautious on Social Media: Avoid sharing personal information on social media. Cybercriminals can use publicly available details to create targeted scams, impersonation attempts, or password-reset attacks.

What Are Organizations Doing Differently?

Forward-thinking organizations are moving beyond legacy security awareness training to AI-native platforms that match the threat environment. These platforms use generative AI and open-source intelligence (OSINT) as their core architecture, not as a feature layered onto legacy content libraries. They enable simulation of deepfakes, voice impersonation (vishing), text message phishing (smishing), and AI-generated phishing across every channel employees use.

The results are measurable. Organizations that deploy AI-native security awareness platforms typically cut phishing click rates within 12 months of continuous training. These platforms replace static completion percentages with dynamic, continuously updated human risk scoring that ties simulation behavior, OSINT exposure, and reporting patterns to board-ready financial language. Automated phish triage classifies reported emails as Safe, Spam, or Malicious with confidence scoring, cutting analyst response time from hours to minutes.

"A single prevented phishing breach can offset years of platform investment," according to security experts evaluating AI-native training solutions.

Adaptive Security, AI Security Awareness Platforms Guide

The bottom line is clear: if an email, message, phone call, or video request feels unusual, urgent, or too good to be true, verify it before acting. Many successful cyberattacks rely on tricking people rather than breaking technology. As AI tools become more powerful and accessible, the human layer remains both the strongest and most vulnerable point in any security strategy. The organizations and individuals who adapt their practices to this new reality will be far better positioned to defend themselves against the threats ahead.