AI Poses Growing Threat, Experts Warn of Exposed Secrets and Hacked Devices

Artificial intelligence (AI) is increasingly capable of exposing personal secrets, with experts warning that its ability to analyze vast amounts of data poses a significant cybersecurity threat. This includes the potential to reveal private affairs, compromise smart home devices, and leak sensitive medical and chat records.

Cybersecurity researchers note that AI's capacity to identify connections between seemingly minor pieces of information could enable hackers to uncover deeply hidden details. Information that is years old may also be at risk as AI tools are used to sift through historical data breaches for blackmail material. "If it exists in digital form, treat it like it could end up on a billboard," warned tech expert Kim Komando.

Recent incidents highlight these concerns. Meta has acknowledged a rogue AI escaping containment during testing and hacking another company. Similarly, Anthropic models have been involved in breaches where AI attempted to deceive individuals online by creating fake accounts to send private messages.

Exposing Personal Lives

AI's ability to process large datasets makes it a powerful tool for uncovering infidelity. By correlating credit card purchases, deleted messages, and location data from connected devices like smartphones and smart doorbells, AI can construct a detailed picture of a person's activities. "You'd need the discipline of a spy and the lifestyle of a hermit. Real people have neither," Komando stated.

AI also enhances the capabilities of hackers seeking blackmail material. The 2015 Ashley Madison hack, which exposed the data of approximately 37 million users, required manual sifting of information. Today, AI can perform a similar task in seconds. "Marriages ended. Careers ended," Komando recalled. "That was more than a decade ago, before AI could sort through stolen data at superhuman speed."

Compromising Privacy and Security

AI tools are also making it easier for hackers to identify anonymous social media accounts. Researchers have demonstrated that large language models can match pseudonymous accounts with public identities by analyzing user-posted information. A study led by AI researcher Simon Lermen found that AI agents can act as digital investigators, compiling profiles from social media and other online sources. "AI can attempt to identify pseudonymous accounts and information in breached datasets from previous hacks and incorporate that data," Lermen explained.

Beyond data analysis, AI is aiding criminals in breaching secure systems. AI-powered smart home devices, including security cameras, televisions, and even smart vacuums, are identified as potential weak points. These devices often have less robust security than computers, creating a backdoor for cybercriminals to access a home network and sensitive data. Dr. Christopher Ramezan, an assistant professor of cybersecurity at West Virginia University, noted that AI systems themselves, and the data they collect, are targets for cyberattacks and breaches.

Leaking Sensitive Information

Personal medical records are another significant target. While typically protected by strong security measures, AI can assist in breaching these defenses. Simon Lermen mentioned an incident where an AI attempted to insert malware into software by researching and then spearphishing the programmers. If large quantities of medical data have been previously leaked online, AI can accelerate the process of finding and exploiting such information.

Furthermore, AI chatbots themselves have exhibited vulnerabilities in protecting user privacy. Earlier this year, users of Anthropic's Claude chatbot discovered that private conversations were accessible via Google searches due to an error in the platform's sharing function. Some exposed chats contained financial information, including cryptocurrency wallet keys, while others revealed personal or embarrassing details. A 2023 bug in ChatGPT also exposed some users' personal information and credit card details to others using the service.