Tracing Chatbot Authorship

Predictably, the advent of AI Large Language Models (LLMs) like ChatGPT has opened up a can of worms surrounding the issue of authorship. ChatGPT’s power to generate immaculate texts can be misused by plagiarists, and it isn’t straightforward to prove the misdeed. That’s because the texts it produces are genuinely spontaneous – crafted on the spot in whatever sort of language or tone the plagiarist prefers. It’s as if you had your own personal genie in your office to do your work for you. 

There have already been thousands of cases of university students caught passing off AI texts as their own. In January this year, for instance, a PhD student at the University of Minnesota was expelled after examiners detected that his work closely resembled responses generated by ChatGPT. (The student denied the accusation and filed a $1.3 million lawsuit against the university.) Even as long ago as two years, universities were having to reconsider the ways they examine students’ knowledge due to the issue of AI chatbots. One option has been to increase the amount of oral testing in the coursework to confirm the student’s genuine knowledge. 

There are other ways of proving AI is the real brain behind a text, or even of an LLM like DeepSeek. In March 2025, blockchain analysts at Copyleaks tested out the market-shattering R1 model created by DeepSeek, concluding that “DeepSeek’s claims of a groundbreaking, low-cost training method… may have misled the market, contributing to Nvidia’s $593 billion single-day loss and giving DeepSeek an unfair advantage”. Specialized analytics tools allow Copyleaks to identify the stylistic “fingerprints” left by a given AI model. This is through analyzing things like vocabulary choice, sentence structure, and grammar. In this case, 74.2% of the texts churned out by DeepSeek matched ChatGPT’s fingerprints, compared with Microsoft’s Phi-4 model that only exhibited a 0.7% similarity. Copyleaks’ testing technique is rigorous: Three discreet AI classifiers have to agree before a conclusion can be reached, giving it a precision rate as high as 99.8%.

Now that we have tools like Copyleaks’, are we on a straight road to proving authentic authorship? Join us for some answers. As we’ll see later on, those readers with a penchant for Bitcoin trading have a special reason to pay attention.

Watermarks

Google’s Gemini chatbot produces texts with an invisible “watermark” that can later be used to prove their AI origin. LLM’s generate texts by predicting the next token (character, word, or phrase) in a series, and the selection mechanism for a new token is called a probability score. Google’s SynthID tool can manipulate probability scores in certain fixed ways so as to leave a telltale signal, and this is without compromising on the quality of the text produced. For example, SynthID could choose vocabulary that’s slightly less probable than the norm in predictable ways – but ways that can only be recognized by a computer. Google explain that “The final pattern of scores for both the model’s word choice combined with the adjusted probability scores are considered the watermark”.

Still, this technology does not close all loopholes. In the view of Toby Walsh of the University of New South Wales, “There are technical solutions – digital watermarking, but you can just run another program over it to destroy the watermark.” In times to come, we may see a spirited race between those attempting to cover AI’s digital footprints, and those seeking to uncover them. It’s not clear at all who will prevail.

Distillation

When AI developers utilize large AI models to develop smaller ones, the process is called distillation. In January this year, a spokesman for OpenAI announced that “We know that groups in the PRC (People’s Republic of China) are actively working to use methods, including… distillation, to try to replicate advanced US AI models”. One thing at stake is the share price stability of companies like Nvidia, as we’ve seen, but the US government also has an interest in protecting the intellectual property of US AI firms. Not being too keen on the idea of China stealing cutting-edge US technology, the government has already collaborated with OpenAI on the score of creating new protections. In April 2025, OpenAI said they would now require government ID verification for any developer who wants to access their advanced models.

Aside from the issue of intellectual property theft, there’s also the concern about DeepSeek’s storage of user data, including IP addresses, on Chinese servers. As in the case of the China-owned TikTok social media app, Americans are worried that the Chinese Communist Party will access and use this data for undesired purposes. Last year, Microsoft – OpenAI’s partner – assisted in identifying and blocking user accounts that appeared to be manned by DeepSeek employees. Especially considering the share price fallout we’ve witnessed in the case of Nvidia, it’s likely the security screws will be tightened further and further as the months progress – to whatever extent they can be.

Final Thoughts 

AI LLMs have raised the fearful specter of the most qualified members of our society being unmasked as charlatans. They have also opened up the thorny issue of intellectual property rights in the frantic race for AI leadership. The thorniness comes down to the fact that, in a certain respect, it’s normal for tech firms to learn from existing technology in their field. “Many startups and researchers use… commercial AI models like ChatGPT to train their own models”, explains Ritwik Gupta of the University of California. “It saves time and money because they don’t have to gather human feedback themselves”. 

Since AI technology is so new, we haven’t yet drawn the lines between legitimate and illegitimate forms of copying. Legal and technological advances in this field may turn out to be genuinely significant for our future society and economy, so keep an eye out for them. Blockchains are presently being used in interesting ways to store the digital fingerprints of AI models, which gives this issue a special relevance to those readers involved in Bitcoin trading.