Researchers have discovered methods to transform CHATGPT and similar AI chatbots into systems for encrypted messaging that evade cybersecurity detection.
A novel approach for embedding cryptography within seemingly human-like messages offers a solution for secure communication in contexts where conventional encryption techniques are easily noticed or restricted, according to team members behind the innovation.
This breakthrough functions like a digital version of invisible ink, rendering true messages visible only to individuals who possess a password or private key. It aims to enhance encrypted communication systems amid increasing hacking incidents and vulnerabilities.
However, researchers cautioned that new cryptographic systems can potentially be misused. They shared their findings in the preprint database on April 11th, arxiv, indicating that it has yet to undergo peer review.
“This research is incredibly promising, but as with any technical framework, ethical considerations arise, highlighting the potential for misuse,” said Mayank Likewar, a researcher in networks and distributed systems at the University of Oslo, in an email to Live Science.
Related: Chinese scientists claim they broke RSA encryption on quantum computers, but there’s a catch
To develop this new encryption technology, researchers implemented a system called embedderllM. This system employs an algorithm to embed secret messages into designated areas of AI-generated text, akin to hidden treasures along a path. The generated text mimics human writing and is undetectable by current decoding methods. The message recipient utilizes a different algorithm that acts like a treasure map to unearth the hidden characters and reveal the message.
Users can send messages crafted by Embedderllm through various messaging platforms, including video game chat platforms, WhatsApp, and more.
“Using LLM for encryption is technically feasible, but it greatly depends on the type of encryption employed,” said Yumin Xia, Chief Technology Officer at Galxe, a blockchain company utilizing established encryption techniques. “Much hinges on the specifics, but this is certainly plausible based on available encryption types.”
The most significant security vulnerability of the method lies at the onset of a message, which involves a secure password exchange to encrypt and decrypt subsequent messages. The system can function with symmetric LLM encryption (where both sender and receiver share a unique private code) or public key LLM encryption (where only the receiver possesses a private key).
Once this key is shared, Embedderllm ensures secure encryption from pre- or post-molecule decryption, and it is built to withstand future advancements in quantum computing and robust decryption methods, the researchers noted.
The team envisions journalists and citizens utilizing this technology to bypass audio restrictions imposed by oppressive regimes.
“We must identify impactful applications for our framework,” stated Reikuwar. “For oppressed individuals, it presents a safer avenue to deliver crucial information without detection.”
He further mentioned that journalists and activists could communicate discreetly in environments of active media surveillance.
Despite significant advancements, experts caution that the practical application of LLM encryption in real-world scenarios is still a way off.
“While some countries have enacted specific restrictions, the long-term significance of this framework ultimately relies on actual demand and adoption,” Xia observed. “As of now, this paper serves as an intriguing exploration of hypothetical use cases.”
Source: www.livescience.com