In Toronto, a quiet revolution is unfolding in a modest office near the University of Toronto, where Biossil, a startup backed by Peter Thiel’s Founders Fund and OpenAI’s startup fund, is using artificial intelligence not to design new drugs from scratch, but to resurrect failed ones—molecules abandoned by Big Pharma due to toxicity, poor bioavailability, or lack of efficacy in initial trials. Earlier this week, the company announced it had successfully repurposed three oncology candidates using its proprietary AI platform, which analyzes decades of failed clinical trial data, genomic profiles and real-world patient outcomes to identify hidden therapeutic windows. This approach could slash drug development timelines by up to 70% and reduce costs from an average of $2.6 billion per new drug to under $800 million, according to internal models shared with Archyde. But beyond the lab, Biossil’s work signals a deeper shift: the globalization of AI-driven pharmaceutical innovation, where intellectual property, regulatory arbitrage, and cross-border data flows are becoming as critical as the science itself.
Here is why that matters: as Western nations tighten export controls on advanced semiconductors and AI models under frameworks like the U.S. CHIPS Act and the EU’s AI Act, Biossil’s model relies on globally distributed data—clinical trial records from India, Brazil, and Nigeria; genomic databases from Estonia and Japan; and real-world evidence from Singapore’s national health system. This creates a new kind of supply chain vulnerability: not of silicon or rare earths, but of ethically sourced, diverse biomedical data. If access to such data becomes politicized—as seen in the 2025 EU-Indian data governance stalemate over clinical trial transparency—startups like Biossil could face bottlenecks that slow innovation just as they begin to scale.
The nut graf is clear: Biossil isn’t just saving failed drugs; it’s building a new paradigm for global health innovation that challenges the traditional West-centric model of drug discovery. By leveraging AI to mine failure, the company turns inefficiency into insight—a stark contrast to the $50 billion annually wasted on abandoned drug candidates worldwide, per a 2024 Nature Reviews Drug Discovery analysis. This isn’t merely a technical upgrade; it’s a geopolitical inflection point. As China accelerates its own AI-driven drug repurposing initiatives through state-backed entities like Insilico Medicine and WuXi AppTec, and as India pushes to develop into a global hub for AI-assisted clinical trials under its National Digital Health Mission, Biossil’s Toronto-London-Singapore operational triangle exemplifies a emerging multipolar innovation network—one where influence flows not from capital alone, but from access to diverse, high-quality data and regulatory agility.
To understand the global stakes, consider this: Biossil’s platform requires continuous input from heterogeneous patient populations to avoid bias in AI training. A 2025 study in Lancet Digital Health found that models trained primarily on Northern European genomic data mispredicted drug response in African and Southeast Asian cohorts by up to 40%. In other words Biossil’s success depends on sustaining partnerships with biobanks in low- and middle-income countries—partnerships that are increasingly scrutinized under new norms of data sovereignty. Earlier this month, South Africa’s Protection of Personal Information Act (POPIA) amendments tightened cross-border health data transfers, requiring explicit re-consent for secondary use—a direct challenge to Biossil’s reliance on legacy trial datasets. Meanwhile, Brazil’s LGPD law now mandates local processing of health data unless equivalent protection is guaranteed abroad, a clause that could force data localization and fragment global AI training sets.
“The real bottleneck in AI-driven drug discovery isn’t compute—it’s consent. If we don’t build trust with Global South communities around how their health data is used, we risk creating AI that works well for some and poorly for others.”
But there is a catch: while Biossil’s model reduces financial risk, it increases regulatory complexity. Unlike traditional drug development, where a single molecule follows a linear path through Phase I-III trials, repurposed drugs often face fragmented regulatory pathways—especially when the new indication differs significantly from the original target. In the U.S., the FDA’s Drug Repurposing Program offers incentives, but lacks binding harmonization with EMA or PMDA (Japan’s agency). This creates a “regulatory arbitrage” risk: startups may prioritize markets with faster approvals, potentially delaying access in regions with greater need. A recent analysis by the Brookings Institution warned that without coordinated international frameworks, AI-driven repurposing could exacerbate therapeutic inequities—where wealthy nations gain early access to revived oncology or neurology drugs, while low-income countries wait years for generic equivalents.
To illustrate the growing divergence in regulatory approaches, consider the following comparison of key jurisdictions’ stances on AI-assisted drug repurposing as of Q1 2026:
| Jurisdiction | AI Drug Repurposing Support | Cross-Border Data Flow Rules | Incentives for Failed Molecule Reuse |
|---|---|---|---|
| United States | FDA’s AI/ML Software as a Medical Device framework; CURES Act provisions | HIPAA permits transfer with de-identification; state-level variations (e.g., CCPA/CPRA) | Orphan Drug Grants; Rare Pediatric Disease Priority Review Vouchers |
| European Union | AI Act classifies most drug discovery AI as “high-risk”; EMA’s BIG initiative | GDPR strict; requires adequacy decisions or SCCs for transfers | Horizon Europe funding; SME Instrument repurposing calls |
| Singapore | Health Sciences Authority (HSA) AI Guideline 2025; National AI Strategy in Health | PDPA allows transfers with Binding Corporate Rules or consent | BIP grant scheme; A*STAR translational funding |
| India | Draft National Strategy for AI in Health (2024); ICMR guidelines | DISHA bill pending; current reliance on IT Rules 2021 and sectoral guidelines | Production Linked Incentive (PLI) scheme for pharma; CDSCO fast-track for repurposing |
This table reveals a critical insight: while all four jurisdictions encourage AI-assisted repurposing, their data governance regimes are diverging—creating compliance overhead that could disproportionately affect startups without global legal teams. Biossil’s decision to incorporate in Canada (with strong AI innovation incentives via the Pan-Canadian AI Strategy) while maintaining legal entities in Luxembourg (for EU data flows) and Singapore (for APAC access) reflects a growing trend: the rise of the “regulatory sandbox multinational”—a startup structured not just for tax efficiency, but for navigational agility across fragmented digital health regimes.
Expert voices confirm this shift is being closely watched by policymakers. In a recent interview with the Financial Times, Pascal Soriot, CEO of AstraZeneca, acknowledged the disruptive potential: “We spent 15 years and over $1 billion on a lung cancer drug that failed in Phase II. If an AI can tell us why it failed in a subset of patients—and how to fix it—we’d be foolish not to listen. But we also need global rules that don’t punish innovation for seeking data where it’s most diverse.”
“The future of medicine isn’t just in Silicon Valley or Basel—it’s in the ability to connect Mumbai’s patient records with Montreal’s AI labs and Osaka’s genomic archives. But that connection requires trust, not just technology.”
Looking ahead, the implications extend far beyond pharmaceuticals. If Biossil’s model proves scalable, it could redefine how nations compete in the 21st-century knowledge economy. No longer will victory belong solely to those with the deepest R&D budgets or the most advanced chip fabs. Instead, advantage will flow to countries that can balance three imperatives: protecting citizen data privacy, enabling ethical cross-border data collaboration, and fostering regulatory environments that reward learning from failure—not just celebrating success. In this new paradigm, Toronto’s quiet AI lab may prove more influential than any summit in Davos or summit in Shenzhen—because it is teaching the world how to listen to the molecules we once threw away.
What do you think: as AI reshapes medicine from a linear pipeline into a circular ecosystem of learning, should global health governance evolve to treat failed clinical trial data not as waste, but as a shared global resource—like the human genome or climate data? And if so, who should steward it?
