The US-Sweden AI Biomedicine Deal Was Signed With Great Fanfare. It Can Be Undone in Six Months.
When Secretary of State Marco Rubio and Swedish Foreign Minister Maria Malmer Stenergard sat down in Helsingborg on May 22 to sign a memorandum of understanding on artificial intelligence and biomedicine, the announcement landed with the usual diplomatic weight: a joint statement, a photo op, and an agreement described in Swedish and American communiques as a landmark in transatlantic technology cooperation.
The document they actually signed says something different. Read the fine print and you find an agreement that commits no funds, creates no legal obligations, and can be dissolved by either government with six months written notice. The headline commitments to joint AI cancer research, antimicrobial resistance cooperation, and health data sharing are, in the documents own language, statements of mutual intent, not promises.
The United States and Sweden have agreed to collaborate on accelerating biomedical innovation using artificial intelligence. The agreement contemplates sharing health data to accelerate drug discovery, coordinating on antimicrobial resistance, and strengthening supply chains for biotech products. It also includes language about reducing reliance on adversary-linked supply chains, a phrase that signals the securitization of biomedical data as a national security concern, not merely a scientific one.
But none of it is enforceable. The MOU explicitly states it does not constitute a legally binding agreement and commits neither government to expenditure of funds. Either party may discontinue participation with 180 days notice. An ongoing collaborative trial between a Swedish university and an American biotech firm, started under this agreement, could be orphaned mid-experiment if diplomatic priorities shift.
This is not unusual for technology prosperity deals of this type. The United States has entered into similar non-binding MOUs with the United Kingdom, Japan, and South Korea, all structured with the same absence of legal or financial commitment. The Swedish deal follows the same template. What it adds is access to Swedens particular assets: the Karolinska Institutet, one of Europes premier medical universities; AstraZenecas Swedish R&D operations; and national health registries that represent some of the worlds most comprehensive longitudinal patient data.
Those assets are real. The commitment protecting them is not.
The research security provisions in the agreement are more durable in practice than the cooperation clauses. The MOU includes provisions on IP safeguards, investment screening, and talent integrity measures tied to shared security concerns, language that will shape how Swedish and American institutions collaborate regardless of whether the broader agreement survives the next administration. A Swedish researcher entering a joint AI drug discovery project with an American counterpart will operate under export control and investment screening frameworks that exist independent of this document.
The biomedicine section of the agreement is where the supply-chain logic is most visible. The document frames AI-enabled drug discovery not as a scientific frontier but as a strategic capability to be secured. Cancer research, rare-disease programs, antimicrobial resistance work, and health data sharing are grouped together with advanced manufacturing and nuclear technology as disciplines requiring bilateral cooperation to reduce dependency on adversary nations. That framing treats patient data and clinical knowledge as inputs to a national technology stack, not as shared human goods.
Sweden accepted this framing. The Swedish governments own press release describes the agreement as involving no legal or financial commitments, a point it emphasizes twice, as if anticipating criticism that the country has given something away. The question of what Sweden gets in return, beyond a photo with the American secretary of state and a place in a network of allied technology partnerships, is not answered by the document.
Six months is not long in drug development. A Phase II trial for an AI-designed oncology compound takes years. A longitudinal health dataset compiled under this agreement would, by the time the first 180-day notice period expired, be just beginning to generate meaningful signal. If the next administration issues that notice, the data partnerships, shared computing infrastructure, and institutional relationships built under the MOU do not disappear, but their legal and political foundation does.
What remains is the underlying science and the people doing it. Swedens researchers will still have Karolinska. American AI companies will still need European health data. The incentive to collaborate will persist even if the document that first formalized it has been filed away. The deals real function may be less about committing either government to a course of action than about establishing the conditions, the security frameworks, the data access norms, the institutional relationships, that make collaboration politically acceptable going forward, regardless of what any single memorandum says.
That is what makes the six-month clause the most honest thing in the document.