AI chatbots can hide secret messages invisible to surveillance

Advancements in artificial intelligence (AI), especially large language models (LLMs) like ChatGPT and open-source counterparts, are reshaping how people communicate daily. These AI systems generate human-like text, answer questions clearly, and even help automate customer service tasks. But now, scientists have found a surprising new use for AI chatbots: securely sending secret messages that can’t be detected by cybersecurity systems.

A recent study, posted on the scientific database arXiv, introduced a novel method that turns AI-generated text into hidden carriers for encrypted messages. This innovative approach offers a crucial alternative for secure communication, especially as governments worldwide seek ways to detect, monitor, or restrict traditional encrypted messages.

Several governments recently pushed for laws that weaken encryption security or mandate backdoors to encrypted data. For instance, the UK government asked Apple to create ways to bypass user encryption. Similarly, France considered laws that would grant authorities access to encrypted communications during investigations. Additionally, Russia-linked hacker groups developed methods to infiltrate supposedly secure platforms like WhatsApp, Signal, and Telegram.

These actions highlight an urgent need for new ways to communicate secretly and safely without traditional encryption, which can now be intercepted or banned. Scientists responded to this challenge by developing an encryption method hidden in AI-generated conversations, making secret messages invisible to cybersecurity and surveillance systems.

Researchers developed a cryptographic method called "EmbedderLLM," a specialized system designed to insert secret, encoded messages within natural-looking text created by AI. This text, appearing completely normal and human-generated, contains hidden messages that only authorized recipients can read using a unique digital "key."

Think of it like writing a message in invisible ink: to the naked eye, the text seems ordinary, but with a special solution (or "key"), the hidden message suddenly becomes clear. The researchers accomplished this by teaching the AI to carefully choose words that not only make sense but also subtly include characters from a hidden, encrypted message at precise intervals. If the AI encounters difficulty in placing these characters naturally, it intelligently reverts and selects alternative words.

Mayank Raikwar, a researcher at the University of Oslo and a key member of the research team, described their method as offering "a safer way to communicate critical information without detection," especially valuable for citizens living under oppressive surveillance.

One remarkable feature of this new encryption approach is that it doesn’t depend on specific AI models. It can operate with any popular chatbot, like ChatGPT or even locally run models, allowing wide accessibility and flexibility.

Furthermore, the technique is resilient against current and future technological threats, including quantum computers. Quantum computing poses a significant threat to traditional encryption because of its potential to easily break codes considered secure today. However, the method developed by Raikwar and his colleagues ensures the encrypted messages remain safe even in the post-quantum era.

"The idea of using large language models for cryptography is technically feasible," said Yumin Xia, chief technology officer at Galxe, a blockchain security company. "While much will depend on the details, this is certainly very possible based on current cryptography."

Despite the exciting possibilities, this new technology comes with ethical concerns. The researchers acknowledge that, like any powerful tool, it could also be misused. "Every technical framework raises ethical issues about misuse, so we must carefully determine where it can safely be applied," explained Raikwar.

Still, potential positive uses abound. The researchers specifically mentioned journalists and activists operating in restrictive environments as groups who could significantly benefit. With oppressive regimes increasingly monitoring or restricting secure communication channels, this AI-driven encryption technique could offer a safe alternative for critical communication.

Though promising, experts caution that widespread adoption may be slow. Yumin Xia noted that while some countries are already imposing encryption restrictions, the broader practical use of this method depends heavily on actual demand and acceptance.

"Right now, the research serves as an interesting exploration of hypothetical scenarios," Xia added, emphasizing that significant real-world implementation might still be some way off.

Nevertheless, Raikwar and his colleagues remain optimistic, hoping their method will soon offer a secure, undetectable messaging solution for those who most urgently need privacy protection.

"Given the increasing risks to data privacy globally," said Raikwar, "finding secure, innovative communication methods is now more important than ever."

As AI continues to advance, the ability to embed hidden messages within AI-generated text could change the landscape of secure communication entirely. This development marks a significant milestone, demonstrating the potential for AI to not just enhance everyday communication but also provide critical protections against surveillance and censorship.

While still early, this breakthrough highlights the evolving role of AI—not merely as a tool for convenience but as a powerful guardian of privacy. As governments and other entities continue to challenge traditional encryption methods, innovative solutions like EmbedderLLM could become crucial tools in protecting freedom of expression and privacy worldwide.

Note: The article above provided above by The Brighter Side of News.

Like these kind of feel good stories? Get The Brighter Side of News' newsletter.