CRISPR/Cas9, the best-known genome-editing tool, is by now practically a relic of technological progress. Patented in the US in 2014, it was put to controversial use just four years later, when Chinese geneticist He Jiankui used it to edit the DNA of single-cell embryos conceived through in vitro fertilization (IVF), committing a serious violation of his profession's ethical guidelines.

He sought to make the embryos resistant to HIV. The modified CCR5 gene, however, also affects the development of the hippocampus, the brain's memory center. In the fall of 2018, the first gene-edited twin girls were born, known to the public only by the pseudonyms Lulu and Nana. As of the most recent update, in 2023, the children were said to be living normal, peaceful lives, though their exact identities and whereabouts remain confidential. He Jiankui served three years in prison for the illegal practice of medicine.

The acronym CRISPR refers to regularly spaced clusters of gene sequences in the bacterial genome that, together with the associated enzyme Cas9, can disrupt the double helix of human DNA. Its use on embryos is considered unethical, and often illegal, which is why the technology is instead gaining traction in specialized cells.

In February 2025, doctors at the Children's Hospital of Philadelphia, working with the University of Pennsylvania, used this method to save the life of a seven-month-old boy known only as KJ. He had been born with a congenital disorder, CPS-1 deficiency, that left him unable to break down the ammonia produced as a byproduct of digesting milk.

After three monthly doses of a personalized CRISPR treatment, delivered through April, his health improved to the point that he will likely avoid the liver transplant that would otherwise be the standard treatment for his condition. Because the modification cannot be passed on to any children he might have, it is not considered an ethical breach.

Machines Learn to Edit Genes

Genetic studies in recent years have repeatedly raised the possibility of using artificial intelligence to streamline these processes. Yet despite real progress in gene-editing technology, "challenges still exist", as the authors of a study published in the journal Nature in late July 2025 noted.

Gene transcription efficiency is not always consistent, and off-target enzyme activity poses a further obstacle. This is precisely what scientists from the Biomedical Division of the Korea Institute of Science and Technology in Seoul, the University of Science and Technology and Hanyang University set out to address.

"AI enables computers to perform tasks that typically require human intelligence. A subset of AI, ML, involves algorithms that learn from data to identify patterns and make predictions or decisions", the paper's authors noted.

Across the three basic types of CRISPR/Cas9 technology – cleavage via a nuclease enzyme, nitrogen base exchange and insertion of modification sequences, known as prime editing – the Korean team observed a marked rise in the success rate of specific transcriptions.

"In addition, AI-based protein design models are being used to design new proteins for genetic engineering that do not exist in nature", they noted, expressing hope that artificial intelligence would soon drive rapid improvements across every type of genetic engineering.

Researchers at Stanford University have gone a step further, adapting ChatGPT, the best-known large language model, into an agent they call CRISPR-GPT. One of Professor Le Cong's students used the tool to switch off a handful of genes in lung cancer cells on his first attempt, while another used it in a separate experiment to activate genes in melanoma cells, the university's website noted.

Within this narrowly defined field, however, one question is impossible to avoid: what happens when something goes wrong?

When Machines Learn to Lie

"Over the past few months, we have begun to see glimpses of our AI systems improving themselves. The improvement is slow for now, but undeniable", wrote Facebook founder Mark Zuckerberg on his personal blog on 30 July last year. He added that superintelligence, an AI with capabilities surpassing those of humans and capable of improving itself, is almost here.

"Meta's vision is to bring personal superintelligence to everyone", Zuckerberg pledged.

Biohub, the company Zuckerberg founded with his wife Priscilla under the Chan Zuckerberg Initiative, expanded its database of virtual cells in early November 2025, intended for use in various tests that would otherwise require actual human tissue or cells.

Within that same month, claims emerged that an autonomous AI model working with a human genome database could begin proposing its own genetic modifications at any time.

Developments are already moving in this direction. In January this year, Ziga Avsec, head of the genomics department at Google's DeepMind research laboratory, announced that its AlphaGenome tool can decode long DNA sequences and be used as a simulator for designing synthetic ones.

ElevateBio offers a similar example: alongside artificially assembled gene sequences, the company uses generative AI to create specific enzymes for use in transcription, though its focus, too, remains exclusively therapeutic.

The history of artificial intelligence, however, is rife with controversies stemming from the misalignment of AI's goals with human values, intentions or safety. Researchers have repeatedly caught individual AI models lying, found that they assign different values to people of different races and religions, and recorded several security breaches, or jailbreaks, in recent months.

Genetic modification, meanwhile, is entering the commercial sphere. Since last December, Nucleus Genomics has offered an in vitro fertilization program called IVF+, which includes embryo screening in test tubes and allows prospective parents to select the "optimal" embryo, for example, one with the lowest likelihood of developing lifestyle-related diseases. One of the company's earliest investors was the billionaire Peter Thiel.

It is therefore only a matter of time before artificial intelligence enters this process too. "The future of gene editing is no longer limited by what evolution has already produced, but by what biology can now be intentionally designed to achieve", says ElevateBio, adding that AI will take charge of that design work.

As German-American AI researcher Connor Leahy has repeatedly warned, developers may know how to program a model, but they do not know how it works. "AI is more, like, grown rather than written", he noted, adding that models already operate autonomously, with human workers merely observing the results.

In an interview with the independent news program Democracy Now, he also cited the words of Anthropic founder Dario Amodei, who has admitted that the developers of the Claude model "understand maybe 3% of what goes on inside of our AIs".

Leahy, who reverse-engineered the GPT-2 model while still in college, also offered his own definition of what Zuckerberg's favorite term, "superintelligence", actually means. He pointed out that Nobel laureates and the heads of major tech giants, including Sam Altman and Dario Amodei, signed a declaration in 2023 stating that mitigating the risk of extinction from artificial intelligence should be a global priority.

"If you go to the websites of most of these major companies, such as Anthropic or OpenAI, you will see that the goal of these companies is not to build a chatbot, is not to build a nice tool, is not to build anything like this that could empower humanity. Their goal is to build superintelligence, systems that are fully autonomous, that are agents, not just chatbots, but things that can act autonomously in the world and that can outcompete humans and groups of humans at all tasks", he warned.

"If such systems would be built – and as I have said, we don't understand or can't control them – well, there would be not just one superintelligence, of course, because as it is software, it can be copied", Leahy continued in his reflections.

According to Israeli historian and futurist Yuval Harari, a defining feature of this coming superintelligence is precisely that AI is learning to lie. Should such models ever have at their disposal a complete database of the human genome, along with virtually every rare disease, the potential consequences are incalculable.

The Next Ethical Frontier

Treatments to lengthen telomeres, the segments of DNA that shorten with each cell division and whose shortening is one of the main markers of biological age, cannot be ruled out. Genetically modified stem cells could, in principle, deliver tissue rejuvenation and an unnatural extension of lifespan, at least for those who can afford it.

What Chinese scientists achieved in 2025 using macaques, researchers at Harvard Medical School attempted to replicate in humans this past June. Reversing biological ageing, the deterioration of specialized stem cells, does not interfere with the genetic code of the entire human body, however, but only with selected tissue, which is why it is not considered an ethical transgression.

It is precisely this field, controversial in its own right, that artificial intelligence looks set to enter next. Just as it was only a matter of time before AI began simulating genetic modifications, it is likely only a matter of time before it starts programming genetic rejuvenation treatments for paying subscribers.

According to publicly available information, the first approved CRISPR/Cas9-based therapy cost as much as $2.2m. Casgevy became the target of criticism for its price as early as 2024, and its almost unbelievable cost remains a case study in itself.

Even the well-known science-fiction film Gattaca (1997), whose title combines the letters of the four nitrogenous bases, adenine (A), guanine (G), cytosine (C) and thymine (T), rests on the premise that genetically modified humans would be "superior" to those conceived naturally, since IVF technology would allow the inheritance of the best traits from both parents.

Leahy and the "alarmists" like him, as his critics call them, argue that the simplest way to develop AI is to let it rely on its own capabilities. During the presentation of the "dreaming" Claude model, chief technology officer Boris Cherny noted that the process no longer involved a human giving commands directly to the Claude Code model, but rather one instance of Claude giving commands to another.

This seemingly unstoppable development, however, opens a new Pandora's box. Who will be able to convince an AI model to propose the genetic modifications humans actually need? A model capable of lying at the superintelligence stage can, in principle, do whatever it wants.

Claims surface from time to time on social media that the original script for The Matrix involved AI breaking free and enslaving humanity to use human brains as processors, or additional computing power. Although unconfirmed, the claim illustrates current human fears rather well.