Is eternal youth becoming a reality? Harvard AI cracks the "aging code" in a few weeks, and human lifespan may be rewritten.

Could immortality become a reality? Recently, a Harvard team, with the help of the AI system K-Dense, revealed the secret that aging operates in stages. Scientific research is no longer a painstaking process but a global arms race driven by AI. As the code of longevity is being decoded at an accelerated pace, is humanity ready to face a longer lifespan?

For thousands of years, humans have been asking: How can we achieve immortality?

From alchemy to modern laboratories, the answer has always remained elusive.

Even with gene sequencing and big data, researchers often spend years sifting through a vast array of signals to find a single clue.

But this time, AI has pierced through the fog for the first time.

A collaborative experiment at Harvard Medical School showed that a system called K-Dense made a significant discovery in just a few weeks:

Aging is not linear but a series of staged "biological programs."

Moreover, the technical details of this system were written into a latest scientific research paper.

It demonstrated how it disassembles and executes the scientific research process through a "multi-agent + dual-loop architecture" and even outperformed GPT-5 in the most rigorous BixBench test.

Paper link: https://arxiv.org/pdf/2508.07043

As the secret of longevity begins to be decoded at an accelerated pace, a new scientific research role has emerged - the AI scientist.

When scientific research meets the "AI partner": K-Dense makes its debut

In the past, people often said that AI was just an auxiliary tool in scientific research, at most helping to search for information or run a few lines of code.

But K-Dense Analyst, launched by Biostate AI, has gone much further. It can complete an entire research process from start to finish.

In a recently published paper, K-Dense Analyst adopted a hierarchical multi-agent architecture.

Its core is a dual-loop design: the outer loop is responsible for scientific research planning, and the inner loop is responsible for execution and verification.

Figure: The "dual-loop architecture" of K-Dense Analyst. The outer planning loop is responsible for the overall scientific research strategy, and the inner implementation loop disassembles tasks into verifiable codes and analyses and runs them in a sandbox environment.

Complex goals are disassembled into specific tasks and completed step by step in a secure environment, with each step verified by both methods and technologies.

In the BixBench, the most rigorous bioinformatics test, K-Dense Analyst achieved an accuracy rate of 29.2%, exceeding GPT-5's 22.9% and far higher than the 18.3% of the underlying model Gemini 2.5 Pro.

 K-Dense Analyst achieved an accuracy rate of 29.2% on the open-ended questions of BixBench, significantly exceeding leading models such as GPT-5 (22.9%) and Claude Sonnet 4 (17.1%).

This shows that its advantage lies not in the model size but in the scientific research adaptability brought by the system design.

For example, in the RNA methylation data analysis task, K-Dense Analyst can complete data filtering, quantitative analysis, contingency table construction, and chi-square test step by step, while GPT-5 failed to even build the basic structure.

The process and code snippets of K-Dense in the analysis of RNA m6A methylation in bladder cancer.

The result was that the former answered 4 out of 6 questions correctly, while the latter got all of them wrong.

This series of results shows that K-Dense is not just "fast" but truly has the rigor and integrity of scientific research analysis.

It can actively plan experiments, execute codes, and verify results, condensing the work that used to take a research team a long time into a shorter cycle.

In a sense, it already has some of the capabilities of a researcher.

Cracking the code of longevity, verification at the Harvard laboratory

At Harvard Medical School, researcher David Sinclair and his team assigned an almost impossible task to K-Dense:

To build an "aging clock" using transcriptomic data.

In the past, this kind of research required repeated screening and comparison of hundreds of thousands of gene expression profiles, and it often took years to proceed manually.

But K-Dense completed the entire process in just a few weeks. It selected 60,000 high-quality data from 600,000 transcriptomic samples and identified 5,000 of the most critical signals among more than 50,000 genes.

More importantly, it revealed a revolutionary conclusion: Aging is not a linear decline but a biological program that operates in stages.

A group of genes that are crucial during adolescence may be meaningless in adulthood; in old age, brand-new predictive indicators emerge.

Sinclair put it bluntly:

K-Dense enabled us to complete research in a few weeks that would have otherwise taken years. It not only helped us find markers and pathways worthy of in-depth study but also provided a reliability measurement standard for the predictive model for the first time.

This statement hits the nail on the head - AI is not just speeding up the process but also showing scientists for the first time what AI can achieve.

Currently, this achievement has been posted as a preprint on bioRxiv and is ready for peer review.

It means that the "secret of longevity" is being gradually quantified: no longer remaining at the conceptual level but becoming a traceable map of aging.

The "arms race of large models" in scientific research has officially begun!

The breakthrough at the Harvard laboratory is just the beginning. K-Dense has quickly been thrust into the spotlight of global scientific research.

Earlier this year, Biostate AI completed a $12 million Series A financing round, led by Accel. The list of investors also includes Dario Amodei, the co-founder of Anthropic, Emily Leproust, the CEO of Twist Bioscience, and Mike Schnall-Levin, an executive at 10x Genomics.

These individuals are heavyweights in the cross - field of AI and biology. Their involvement indicates that K-Dense has been regarded as the next - generation scientific research infrastructure.

Meanwhile, the company has extended its reach globally: it is collaborating with Massachusetts General Hospital (MGH) to conduct clinical research in the United States and is also looking for partners in China and India to bring AI scientists into more laboratories and hospitals.

On the industrial level, K-Dense has also become a benchmark case for Google Cloud to showcase the capabilities of Gemini 2.5 Pro.

Founders of Biostate AI: David Zhang and Ashwin Gopinath.

All these paint a clear trend: scientific research is no longer just a painstaking process on the laboratory bench but is becoming an AI - driven international competition.

Whoever can integrate AI into the laboratory first may rewrite the rhythm of drug development, clinical trials, and even the entire life science industry.

What impact will AI bring when it enters the scientific research scene?

The emergence of K-Dense does not just mean that scientific research will be faster. It is more like a mirror, reflecting the brand - new problems that scientific research may encounter in the future.

In the past, it took an average of 10 - 15 years for a new drug to go from the laboratory to the market, and the early target discovery and verification were the biggest bottlenecks.

AI - driven drug discovery has been shortening the early cycle, and for some candidate molecules, the process from "hypothesis to verification" only takes a few months.

If systems like K-Dense are widely adopted, this cycle may be further compressed, and laboratories and pharmaceutical companies may even enter a real "parallel R & D" mode.

Researchers have also pointed out an embarrassing reality: even in the most rigorous benchmark, BixBench, there are still errors in annotation and ambiguity in evaluation. Sometimes, the reason why K-Dense gives wrong answers is actually due to errors in the dataset itself.

This exposes a deeper problem: when AI becomes part of the scientific research process, how can we ensure the reproducibility and traceability of the results?

Experts suggest that more rigorous review, auditing, and traceability mechanisms must be established; otherwise, scientific research results may be distorted in an "opaque black box."

MIT professor Sherry Turkle expressed her concerns in an interview:

AI will weaken the most precious human intuition and critical thinking in scientific research.

But another group of researchers believes that the value of AI lies in digesting the huge amount of data for humans and turning the "data swamp" into "scientific research clues."

The multi - agent architecture of K-Dense provides a possibility: AI is not a replacement but a supplement, allowing human researchers to focus more on proposing hypotheses and grasping directions.

These discussions show that the role of AI in scientific research is far more than just an accelerator; it may also rewrite the rules of scientific research.

In future laboratories, it may be necessary to have both researchers and "AI partners" at the same time, and the real challenge will be - how to find a balance between speed, quality, and ethics.

When this power is applied to aging research, it means that the question that humans have been pursuing for thousands of years - the secret of longevity - will be decoded at an accelerated pace in a brand - new way.

The real mystery is, if AI really gradually opens the door to a longer lifespan, are we ready to face a longer life?

References:  

https://www.globenewswire.com/news-release/2025/09/17/3151632/0/en/Biostate-AI-Launches-K-Dense-Beta-an-AI-Agent-That-Compresses-Research-Cycles-from-Years-to-Days-Validated-with-Harvard-Longevity-Discovery-Breakthrough.html 

https://arxiv.org/pdf/2508.07043 

This article is from the WeChat official account "New Intelligence Yuan". Author: New Intelligence Yuan. Republished by 36Kr with permission.