AI Drug Discovery Reaches the Phase 3 Epoch: Inside the September 2026 Inflection Point

For nearly a decade, the core question haunting artificial intelligence in drug discovery has been theoretical: can computational platforms design novel molecules that actually survive the crucible of human clinical trials?

In September 2026, the industry officially crossed from theoretical promise into registrational reality. On September 9, 2026, the first patient was dosed at Peking Union Medical College Hospital in GENESIS-IPF-3 — the global Phase 3 clinical trial of Insilico Medicine’s rentosertib, making it the first drug in history where both the therapeutic target and the molecular chemical structure were discovered and designed by AI to enter registrational Phase 3 testing. Exactly one week later, on September 16, 2026, Novo Nordisk announced a multi-year strategic alliance with Anthropic to deploy Claude Science across its discovery pipeline, backed by Anthropic’s new wet-lab validation infrastructure and its $400M acquisition of Coefficient Bio.

Between these clinical milestones, Generate:Biomedicines is advancing dual Phase 3 trials for an AI-designed monoclonal antibody, Recursion Pharmaceuticals is scaling its phenomics platform with multiple clinical assets and big-pharma partnerships, and Xaira Therapeutics is redefining target discovery through genome-scale virtual cell models.

The narrative of “AI as a magical shortcut” has evaporated. In its place stands something far more durable: dependable, production-scale infrastructure meeting the unforgiving reality of human biology.

---

The Clinical-Stage AI Cohort: September 2026 State of Play

While the broader AI-bio directory lists over 330 companies, only a select cohort possesses clinical-stage assets designed computationally from target identification or de novo molecular generation. Below is the current landscape of leading AI-derived programs in active clinical development:

Molecule / Program	Developer	AI Platform / Modality	Target & Indication	Clinical Phase	Key Pharma Partner
Rentosertib (ISM001-055)	Insilico Medicine (HKEX: 3696)	Pharma.AI (PandaOmics + Chemistry42); Small molecule	TNIK; Idiopathic Pulmonary Fibrosis (IPF)	Phase 3 (GENESIS-IPF-3, first patient dosed Sept 9, 2026)	Lilly ($2.75B pact), Servier ($888M), SK Biopharm
GB-0895 (golukibart)	Generate:Biomedicines (NASDAQ: GENB)	Chroma generative protein model; De novo antibody	TSLP; Severe Asthma & COPD	Phase 3 (SOLAIRIA-1 & SOLAIRIA-2)	Novartis ($1B+ biobucks), Pfizer
AU-007	Biolojic Design	Computational epitope & switch design; Monoclonal antibody	Human IL-2 (conditional binder); Unresectable solid tumors	Phase 2 (First AI-designed antibody to enter Phase 2)	Merck KGaA ($376M)
REC-1245	Recursion Pharmaceuticals (NASDAQ: RXRX)	Recursion OS (Phenomics + ML-guided medicinal chemistry)	RBM39 (RNA splicing degrader); Solid tumors & Lymphoma	Phase 1/2 (Target to IND in <18 months)	Roche / Genentech, Sanofi
IAM1363	Iambic Therapeutics	NeuralPLexer physics-informed neural network; Small molecule	Brain-penetrant HER2 inhibitor; HER2-amplified solid tumors	Phase 1/1b	Takeda ($1.7B biobucks)
GB-4362	Generate:Biomedicines (NASDAQ: GENB)	Generative protein design; MMAE neutralizer	Reversal of ADC toxicity; Oncology supportive care	Phase 1 (Granted FDA Fast Track in 2026)	Wholly owned
REC-7735	Recursion Pharmaceuticals (NASDAQ: RXRX)	AI-guided medicinal chemistry; Small molecule	Mutant-selective PI3Kα inhibitor; Solid tumors	Phase 1/2 (Cleared IND H2 2026)	Sanofi
Isomorphic Small Molecules	Isomorphic Labs (Alphabet)	AlphaFold 3 / Generative small-molecule engines	Undisclosed oncology & immunology targets	IND-Enabling (First clinical filings targeted late 2026)	Novartis ($1.2B+), Eli Lilly ($1.7B+), J&J

---

1. Insilico Medicine: Rentosertib Enters Phase 3 (GENESIS-IPF-3)

The single most consequential clinical event for AI drug discovery in 2026 occurred on September 9, when the first patient was officially dosed in the GENESIS-IPF-3 trial (NCT07687459 / CTR20262475) at Peking Union Medical College Hospital, with Shanghai Pulmonary Hospital initiating concurrent enrollment.

[Target Identification]        [Generative Chemistry]          [Phase 1 / 2a Safety & Signal]       [Phase 3 Registrational]
      PandaOmics            →       Chemistry42             →         GENESIS-IPF (N=71)          →      GENESIS-IPF-3 (N=320)
Identified TNIK as novel      Generated <80 molecules;        Met primary safety endpoint;          Randomized, double-blind,
fibrosis & aging target       Selected ISM001-055 in 18 mo    +98.4 mL FVC signal at 60mg QD         52-week FVC decline across 47 sites

The Trial Architecture

• Design: Prospective, multi-center, randomized, double-blind, placebo-controlled study across 47 specialized pulmonology centers in China.
• Enrollment: 320 patients with confirmed idiopathic pulmonary fibrosis (IPF).
• Primary Endpoint: Absolute annual rate of decline in Forced Vital Capacity (FVC, mL/year) over 52 weeks — the gold standard registrational endpoint established by regulatory agencies for pirfenidone and nintedanib.
• Secondary Endpoints: Disease progression-free survival, time to first acute IPF exacerbation, King’s Brief Interstitial Lung Disease (K-BILD) health-related quality-of-life scores, and long-term safety/tolerability.

The Scientific Significance

Rentosertib represents an entirely unprecedented milestone in pharmaceutical history:

1. Target Novelty: Unlike “fast-follower” AI efforts that design better binders to known, de-risked targets (e.g., EGFR or KRAS), TNIK (TRAF2- and NCK-interacting kinase) had never previously been targeted for pulmonary fibrosis in clinical medicine. It was discovered by Insilico’s PandaOmics pathway engine analyzing multi-omics datasets across aging and fibrotic tissues.
2. De Novo Generation: The chemical structure was designed from scratch by Chemistry42 without starting from an existing chemical scaffold. The program required synthesizing fewer than 80 molecules before nominating the clinical candidate.
3. Biological Age Reversal Context: Just days prior to Phase 3 dosing, in early September 2026, a landmark paper appeared in Nature Biotechnology evaluating secondary plasma proteomics from the Phase 2a trial. The analysis revealed that rentosertib administration was associated with a statistically significant reversal in predicted biological age across six validated proteomic aging clocks — providing molecular evidence for the dual anti-fibrotic and anti-aging mechanism of TNIK inhibition.

If GENESIS-IPF-3 demonstrates statistically significant deceleration in FVC decline with acceptable hepatic tolerability when data unblinds (~2028), it will irrevocably validate the commercial thesis that an algorithm can discover novel biology and deliver an approvable drug.

---

2. Big Tech Goes Vertical: Novo Nordisk, Anthropic, and the ‘Claude Science’ Era

If 2024 and 2025 were defined by big pharma licensing point-solution AI software, late 2026 is defined by deep structural alliances where frontier AI laboratories integrate directly into pharma discovery pipelines.

On September 16, 2026, Danish pharmaceutical powerhouse Novo Nordisk signed a landmark enterprise collaboration with Anthropic to accelerate cardiometabolic and rare disease discovery using Claude Science, Anthropic’s domain-specialized life-sciences reasoning engine.

Dimension	Anthropic + Novo Nordisk (Sept 2026)	Google Cloud + Merck (2026)	Isomorphic Labs + Lilly / Novartis
Deal Structure	Strategic AI enterprise co-development	$1B multi-year cloud/enterprise AI pact	Multi-target discovery partnership ($3B+ combined)
Core Technology	Claude Science reasoning models + Agentic workflows	Gemini Enterprise across R&D & manufacturing	AlphaFold 3 + proprietary generative molecular chemistry
Laboratory Loop	Anthropic internal wet-lab verification + Novo labs	Merck enterprise laboratories	Fully automated robotic wet labs (London)
Target Scope	Cardiometabolic, obesity, metabolic dysfunction, rare disease	Global portfolio cross-modality R&D	Specific high-value oncology & immunology targets

Anthropic’s Strategic Transformation

The Novo Nordisk partnership highlights Anthropic’s calculated pivot into life sciences throughout 2026:

• Coefficient Bio Acquisition ($400M, April 2026): Anthropic acquired biotech startup Coefficient Bio, gaining internal biological domain expertise and proprietary wet-lab assay facilities to build closed-loop reinforcement learning pipelines.
• Claude Science Reasoning Platform: Unlike general-purpose language models, Claude Science is tuned specifically on structural biology, biochemical reaction pathways, stoichiometric constraints, and clinical trial data.
• Autonomous “AI Scientist” Agents: Research published earlier in 2026 demonstrated that multi-agent LLM systems could independently formulate antibody-drug conjugate (ADC) linker and target strategies that mirrored solutions previously identified by elite human medicinal chemistry teams. Novo Nordisk aims to deploy these agents to automate initial target validation and pharmacokinetic hypothesis testing.

---

3. Generative Protein Biology: Generate:Biomedicines in Registrational Trials

While small molecules led the initial AI-bio wave, generative protein design has matured at extraordinary speed. Generate:Biomedicines (NASDAQ: GENB), which completed a successful initial public offering in February 2026, is demonstrating that computationally generated macromolecules can rival or exceed traditional antibody discovery technologies like phage display and transgenic mice.

The SOLAIRIA Phase 3 Program

Generate is currently evaluating its lead asset, GB-0895 (golukibart), across two global Phase 3 trials:

• SOLAIRIA-1 & SOLAIRIA-2: Double-blind, placebo-controlled trials testing GB-0895 in patients with severe, uncontrolled asthma regardless of baseline eosinophil counts.
• Target: Thymic Stromal Lymphopoietin (TSLP), an upstream epithelial cytokine that triggers multiple downstream inflammatory cascades in airway disease.
• Technical Origin: Designed de novo using Chroma, Generate’s foundation protein model. Chroma generates entire tertiary protein backbones conditioned on targeted binding pockets and biophysical constraints (solubility, thermal stability, low immunogenicity).
• Expansion: Concurrently advancing a Phase 1b trial in Chronic Obstructive Pulmonary Disease (COPD).

In addition to GB-0895, Generate received FDA Fast Track designation in mid-2026 for GB-4362, an engineered protein designed to neutralize free monomethyl auristatin E (MMAE) in circulation. MMAE is a potent cytotoxic payload used in multiple commercial antibody-drug conjugates (ADCs); GB-4362 acts as a circulating “safety sponge” to mitigate systemic off-target toxicities without compromising tumor killing.

---

4. Industrial Phenomics & Degrader Chemistry: Recursion OS

Recursion Pharmaceuticals (NASDAQ: RXRX) enters autumn 2026 in the midst of an aggressive operational transition from high-burn exploratory platform to disciplined clinical-stage execution. Following its $688M acquisition of Exscientia, Recursion unified its 50+ petabyte phenomics imaging database with Exscientia’s AI-guided precision chemistry.

Pipeline Highlights

• REC-1245 (RBM39 Degrader): Recursion’s first fully AI-discovered asset is advancing through Phase 1/2 clinical trials in biomarker-enriched solid tumors and lymphoma. The program was discovered using high-content microscopy and computer vision on disease-perturbed human cell lines, progressing from novel target nomination to IND-enabling studies in under 18 months.
• REC-7735 (PI3Kα Mutant-Selective): Received U.S. FDA IND clearance in mid-2026. The molecule was engineered computationally to selectively inhibit oncogenic PI3Kα mutants (H1047R and E545K) while sparing wild-type PI3Kα, aiming to eliminate the severe hyperglycemia that has historically plagued the PI3K inhibitor class.
• REC-4881 (MEK1/2 Inhibitor): Lead program for Familial Adenomatous Polyposis (FAP), demonstrating sustained polyp regression in Phase 2, with detailed registry data scheduled for presentation in November 2026.
• Big Pharma Validation: Genentech recently advanced the first neuroscience target nominated under their multi-target collaboration into formal pipeline development, unlocking milestone payments and validating Recursion’s automated wet-lab phenomics engine.

---

5. Next-Generation Causal Models: Xaira Therapeutics and X-Cell

In April 2024, Xaira Therapeutics launched with $1 billion in committed capital backed by ARCH Venture Partners, Foresite Capital, and F-Prime, under the leadership of Dr. Marc Tessier-Lavigne. In 2026, Xaira began demonstrating what that capital base was built to achieve.

In September 2026, Xaira was awarded the 2026 Fierce AI Innovation Award for Preclinical Development, following the launch of X-Cell, its foundation virtual cell model.

+-------------------------------------------------------------------------------+
|                           X-ATLAS / PISCES DATASET                            |
|             Genome-scale, multi-modal single-cell perturbation maps            |
+---------------------------------------+---------------------------------------+
                                        |
                                        v
+-------------------------------------------------------------------------------+
|                                 X-CELL MODEL                                  |
|         Predicts how cell states respond to targeted genetic/chemical hits     |
+---------------------------------------+---------------------------------------+
                                        |
                 +----------------------+----------------------+
                 |                                             |
                 v                                             v
[Causal Target Identification]             [De Novo Biologics Design]
Identifies true disease drivers            Generates antibodies targeting functional
rather than downstream epiphenomena        epitopes mapped by in silico perturbation

By moving from static structure prediction (predicting how a single protein folds) to dynamic systems perturbation (predicting how a living cell responds to a genetic or chemical perturbation), Xaira represents the third wave of AI in biology: Causal Biology.

---

6. The Reality Check: “Biology is Kingmaker” and Regulatory Harmonization

Despite the surge in clinical assets, the prevailing consensus among pharmaceutical chief scientific officers in late 2026 has sobered: AI cannot rewrite the laws of human physiology.

A widely cited 2026 perspective in Nature Reviews Drug Discovery emphasized that while AI tools compress the exploratory target-to-candidate timeline from 4.5 years down to 12–18 months, the clinical failure rate of early-stage assets remains largely tethered to biology. Computational affinity does not guarantee oral bioavailability, predictable human metabolism, low immunogenicity, or lack of idiosyncratic organ toxicity.

[!IMPORTANT] The Regulatory Standard Has Not Changed: Both the FDA and EMA have formally reaffirmed that AI-derived drugs will face the exact same evidentiary standards as chemically synthesized or cell-line-derived molecules. In early 2026, the FDA and EMA jointly released the Guiding Principles of Good AI Practice in Drug Development, establishing strict transparency, reproducibility, and auditability requirements for computational pipelines used in regulatory filings.

Key Tenets of the 2026 FDA/EMA Guiding Principles:

1. Audit Trails for Model Training: Sponsors must document training data provenance, data splits, and leakage prevention for any algorithm used to justify dosing, target selection, or patient stratification.
2. Mechanistic Orthogonality: Computational predictions cannot serve as standalone proof of mechanism; sponsors must provide orthogonal wet-lab biochemical and phenotypic validation.
3. Generalizability Testing: Models trained on historical molecular libraries must demonstrate predictive validity on novel chemical scaffolds outside the training distribution.

---

The 2026–2027 Catalyst Calendar

The next twelve months will provide the first definitive empirical readouts on whether AI-designed therapeutics can clear Phase 2 and Phase 3 efficacy thresholds:

Timeframe	Company	Asset	Event / Catalyst	Field Impact
Q4 2026	Isomorphic Labs	Undisclosed	First IND submission for AlphaFold-3-derived small molecule	First clinical test of DeepMind / Isomorphic design platform
Q4 2026	Recursion	REC-4881	Presentation of complete Phase 2 FAP polyp burden data	Proof of phenotypic screening translation
H1 2027	Generate:Biomedicines	GB-0895	Interim Phase 3 SOLAIRIA-1 data in severe asthma	First pivotal Phase 3 readout for a de novo generative antibody
H1 2027	Biolojic Design	AU-007	Updated Phase 2 overall response rate data in solid tumors	Validation of computationally engineered smart switch antibodies
H2 2027	Iambic Therapeutics	IAM1363	Phase 1/1b intracranial efficacy readout in HER2+ brain metastases	Validation of physics-informed AI for blood-brain-barrier penetration
2028	Insilico Medicine	Rentosertib	Registrational Phase 3 GENESIS-IPF-3 primary endpoint readout (FVC decline)	Definitive historical test for end-to-end AI small molecule discovery

---

The Bottom Line

The AI biology landscape in September 2026 is no longer defined by who can train the largest transformer on UniProt or who can generate the most dazzling 3D renderings of binding pockets.

The industry has stratified into two fundamentally distinct tiers:

1. The Registrational Cohort: Companies like Insilico Medicine, Generate:Biomedicines, Recursion, and Biolojic Design that have pushed computationally designed molecules into multi-center Phase 2 and Phase 3 human clinical trials, validated by multi-billion-dollar big-pharma counter-parties and sovereign balance sheets.
2. The Exploration Long Tail: Several hundred early-stage startups that possess algorithmic platforms but lack clinical-grade pipelines, internal wet-lab verification loops, or sufficient capital to survive the ongoing biotech financing consolidation.

With the first patient now dosed in Phase 3 for rentosertib, and with AI-designed antibodies in registration-directed trials, the era of speculation is officially over. The clinical epoch of AI biology has begun.

---

Sources & Further Reading:

• Insilico Medicine Announces First Patient Dosed in Phase 3 GENESIS-IPF-3 Clinical Trial (Press Release, September 9, 2026; CTR20262475 / NCT07687459).
• Proteomic Clock Analysis of Rentosertib in Idiopathic Pulmonary Fibrosis, Nature Biotechnology (September 2026).
• Novo Nordisk and Anthropic Form Strategic Collaboration to Accelerate Drug Discovery with Claude Science (Novo Nordisk Corporate Press Release, September 16, 2026).
• Xu, Z., Ren, F., Wang, Q. et al. A small-molecule TNIK inhibitor for idiopathic pulmonary fibrosis: a randomized, double-blind, placebo-controlled Phase 2a trial. Nature Medicine 31, 2602–2610 (2025).
• Generate:Biomedicines Corporate Update and SOLAIRIA Phase 3 Trial Progress (August–September 2026; NASDAQ: GENB).
• Recursion Pharmaceuticals Reports Q2 2026 Pipeline Advancements and Operational Discipline (August 2026; NASDAQ: RXRX).
• FDA & European Medicines Agency: Guiding Principles of Good AI Practice in Drug Development (Joint Regulatory White Paper, 2026).
• Fierce Biotech & Fierce AI Innovation Awards 2026: Preclinical Development and Discovery Categories.