While artificial intelligence companies claim AI could help discover cures for most human diseases within a decade, Northeastern University graduate students have found a starkly different reality: today's AI models make dangerous factual mistakes, cannot draft credible scientific reports independently, and require constant human guidance to produce usable results.

The AI drug discovery market was valued at $2.3 billion in 2025 and is projected to reach $13.8 billion by 2033, according to market analysis firm Grand View Research. Yet behind the billion-dollar investments and optimistic headlines lies a gap between marketing claims and what these tools can actually accomplish in real pharmaceutical research.

What Did Northeastern Students Discover About AI Drug Discovery Tools?

Over the past academic year, three cohorts of graduate students in Anton Sinitskiy's Applied AI capstone course at Northeastern tested some of the most popular open-source AI frameworks used for complex medical research, including GPT Researcher and Agent Laboratory. Their findings, published across three separate scientific reports between January and June 2026, paint a sobering picture of AI's current limitations in pharmaceutical research.

In their first report, shared in January 2026, students analyzed eight open-source AI models to see if they could reproduce human-made algorithms designed for drug discovery. One algorithm they tested was developed at Northeastern by researchers working under Lei Xie, a professor of pharmaceutical and biomedical sciences. This algorithm is useful for predicting how molecules interact during drug development. The result: all eight open-source models failed to even draft scientific reports and were incapable of reproducing the algorithm.

For the second report, released on June 27, students tested five more advanced AI frameworks and asked them to replicate findings from a drug discovery study developed by Swiss pharmaceutical company Novartis. While these more advanced models showed slight improvement in drafting original hypotheses and full reports, the work lacked depth and contained numerous errors.

In the most recent report published on June 29, students tested whether AI models could independently review scientific papers, analyze large datasets, and evaluate formulas. Once again, the models largely failed, producing inaccurate and misleading results.

How Can AI Actually Help Drug Discovery Right Now?

The Northeastern research wasn't entirely negative. Students discovered that AI does have genuine value in specific, limited contexts. When researchers worked collaboratively with AI models rather than relying on them to work independently, the results improved dramatically. In simple cases, AI can accomplish about 95 percent of a task on its own, but in more complex scenarios, it can fail badly and requires what researchers call "hand holding".

• Repetitive Tasks: AI excels at automating routine, standardized work that doesn't require creative problem-solving or novel scientific judgment.
• Statistical Support: AI can assist with data analysis and statistical calculations when human researchers verify the outputs and catch errors.
• Stress Testing Ideas: AI can help researchers explore variations of hypotheses and test multiple scenarios quickly, though humans must evaluate the scientific validity of results.

"If you just listen to what is written in the media, on scientific papers, and marketing materials, it sounds like AI is capable of solving all problems, including drug design. But unfortunately the truth is very far from these overhyped statements," said Anton Sinitskiy, a teaching professor in the Applied AI program at Northeastern.

Anton Sinitskiy, Teaching Professor, Applied AI Program, Northeastern University

Lei Xie, the Northeastern professor whose algorithm the students tested, emphasized that the findings put into perspective "the scope" of these technologies. While AI can be useful in some contexts, it cannot "replace human intelligence".

Why Are Today's AI Models Unreliable for Drug Discovery?

The core problem isn't with any single AI framework but with the current state of the field itself, according to Sinitskiy. Today's large language models, which power most AI drug discovery tools, are prone to "hallucinations," a technical term meaning they confidently generate false information that sounds plausible. In pharmaceutical research, where accuracy is literally a matter of life and death, this flaw is particularly dangerous.

These models also struggle with the kind of rigorous, source-backed scientific reasoning that drug discovery demands. They cannot reliably cite their sources, verify claims against existing literature, or acknowledge the limits of their knowledge. For a field where every claim must be traceable and verifiable, this represents a fundamental limitation.

Niventhini Senthilselvan, a student co-author of one of the reports who now works as an AI engineer in the telecommunications industry, reflected on what the research taught her about using AI responsibly. "It taught me to look beyond simply getting an answer from AI and ask deeper questions: How reliable is it? Why did the model produce it? And how can we improve it?".

"As AI becomes increasingly integrated into real-world decision-making, understanding not just an answer, but its reliability and uncertainty, is equally important," Senthilselvan noted.

Niventhini Senthilselvan, AI Engineer

What Does This Mean for the Booming AI Pharma Industry?

Despite the Northeastern findings, the commercial AI drug discovery sector continues to grow rapidly. Insilico Medicine, a clinical-stage AI-driven drug discovery company, reported total revenue of $106.3 million in the first half of 2026, representing a 287 percent year-over-year increase. The company achieved a net profit of $35.54 million during the same period, with a gross profit margin of 90.3 percent.

Much of this revenue growth came from large upfront payments and milestone payments from major pharmaceutical partnerships. In March 2026, Insilico Medicine signed a collaboration agreement with Eli Lilly valued at up to $2.75 billion, including a $115 million upfront payment. In June, the company announced a $2.5 billion AI drug discovery collaboration with SK Biopharmaceuticals focused on neuroimmune diseases.

These partnerships suggest that major pharmaceutical companies view AI as a valuable tool within their broader research operations, even if AI cannot yet operate independently. The collaborations often involve Insilico's software platform being used alongside traditional pharmaceutical expertise, which aligns with the Northeastern students' finding that AI works best when humans remain actively involved.

Sinitskiy emphasized that the goal of the Northeastern research was not to condemn AI drug discovery entirely, but to establish realistic expectations. "The point is AI can do really important things, but you need to know what kind of things it can do, and you need to run it properly," he said.

Sinitskiy

As the AI drug discovery market continues its rapid expansion, the Northeastern findings serve as a crucial reality check. The technology shows genuine promise in specific applications, but the path to transforming pharmaceutical research requires honest assessment of current limitations, not marketing hyperbole about near-miraculous breakthroughs.