Cross-Border
Multi-jurisdiction IP Transfer for AI Research Lab
$35M
IP Value
4
Jurisdictions
23
Patent Families
10 weeks
Timeline
The Challenge
What Our Client Faced
An AI research lab based in Singapore with patent filings in Singapore, the US, the EU, and China was acquired by a US technology company. The IP portfolio consisted of 23 patent families covering novel architectures for large language models, efficient training techniques, and inference optimization methods. The transfer involved several unique complexities: 7 patent applications were still pending and in prosecution, requiring careful coordination with patent offices to ensure continuity; 4 patents had been co-invented by researchers who had since left the company, and their assignments needed to be confirmed; the lab's trade secrets — including proprietary training data, model weights, and unpublished research — represented significant value but were difficult to define and transfer under traditional IP transfer frameworks; cross-border IP transfer triggered potential withholding tax obligations under Singapore's tax treaties with the US, the EU, and China; and certain patent filings in China were subject to export control review under China's revised Patent Law provisions on state-secret-related inventions.
Our Approach
How We Handled It
We developed a comprehensive IP transfer framework that addressed both registered and unregistered IP assets. For the 23 patent families, we prepared and filed assignment documents with all four patent offices — IPOS (Singapore), USPTO (US), EPO (EU), and CNIPA (China). The 7 pending applications required additional coordination with the prosecuting patent attorneys to ensure that the assignments did not affect prosecution timelines. We confirmed assignments from the 4 former researchers through a combination of original agreement enforcement (where the agreements were clear) and retroactive assignment execution (where gaps existed). Two former researchers required negotiated settlements involving nominal consideration and mutual releases. For the trade secrets, we developed a novel trade-secret transfer protocol that included: a comprehensive inventory of all trade secret assets (training data, model weights, hyperparameter configurations, unpublished research papers, and internal documentation); a formal chain-of-custody transfer process with documented handoffs; technical access controls to ensure that trade secrets were accessible only to authorized personnel during the transition; and post-transfer confidentiality obligations binding both the seller's remaining personnel and the buyer's receiving team. For the tax structuring, we worked with tax counsel in each jurisdiction to optimize the IP transfer structure. We structured the transfer through Singapore's IP holding company, utilizing Singapore's Section 19B capital allowances for IP acquisition and the Singapore-US tax treaty's reduced withholding rate on royalties. The China patent transfer required filing with the State Intellectual Property Office and obtaining clearance under the state-secret review process, which we initiated in parallel with the other workstreams.
The Outcome
Results & Impact
All 23 patent families were successfully transferred across all four jurisdictions within 10 weeks. The pending patent applications continued prosecution without interruption. All former-researcher assignments were confirmed or executed, closing the chain-of-title gaps. The trade-secret transfer protocol we developed was described by the acquirer's CTO as 'the most comprehensive IP handoff process we have ever seen' — it became the template for the acquirer's subsequent AI research acquisitions. The tax structure saved the parties an estimated $2.8M in withholding taxes compared to a naive transfer approach. The China clearance was obtained in 7 weeks, within our projected timeline. Post-closing, we provided 90 days of IP integration support, managing the final patent office recordings and monitoring the pending applications through their next prosecution milestones.
Key Takeaways
Lessons Learned
- Trade secrets in AI — including model weights and training data — require formal transfer protocols, not just contractual assignment language.
- Pending patent applications require special coordination during transfers to avoid prosecution delays.
- Former-employee IP assignments should be confirmed proactively; gaps are common and become expensive to fix during a transaction.
- Cross-border IP transfers have significant tax implications that can be optimized through proper structuring.
- China patent transfers require state-secret clearance that should be initiated early in parallel with other workstreams.