On June 12th, at 5:21 PM ET, the U.S. government issued a sudden directive to Anthropic, an artificial intelligence firm, mandating the suspension of access to its advanced models, Fable 5 and Mythos 5, for all foreign entities—including its own international staff. The directive came without prior warning or a timeline for future access.
Operational dependencies on these AI models were minimal, largely due to their recent availability. Fable 5 was released just a week prior, while Mythos 5 had been accessible to a limited number of partners as part of Anthropic’s Project Glasswing. However, by July 1, the U.S. government lifted the suspension, restoring global access to these models.
This brief disruption sent a clear signal to governments worldwide about the precarious nature of frontier AI access, which can be abruptly curtailed without notice or recourse.
National security was cited as the official rationale for the suspension. Anthropic suggested that the government had identified a security vulnerability in Fable 5, which the company characterized as a misunderstanding, cautioning that such stringent measures could significantly impede industry innovation. Additionally, this incident surfaced against the backdrop of Anthropic’s recently rejected Pentagon contract that would have allowed military utilization of its technologies for numerous purposes, including autonomous systems. Following the breakdown of negotiations, the Department of Defense labeled Anthropic a “supply chain risk to national security”—a classification typically reserved for foreign adversaries and unprecedented for a domestic company. Anthropic has initiated legal action to contest this designation, highlighting the already contentious relationship with the U.S. government.
African governments found themselves blindsided by this development, learning of the suspension through media reports just three days after the directive was issued.
This incident underscores a larger issue: Africa’s significant reliance on foreign technology, evident in critical infrastructure sectors such as telecommunications, cloud computing, and digital identity systems dominated by firms from the U.S., Europe, and China. The June suspension not only affected international institutions but also restricted access for some of Anthropic’s U.S.-based staff, amplifying the realization that the power to sever lines of technological engagement lies solely with foreign governments.
The directive aligns with the U.S. AI Action Plan, which emphasizes assessing national security risks associated with frontier technologies. This raises a pressing question: how will African institutions address their reliance on technology that could be revoked at the discretion of the U.S. government?
Historical Precedents
This is not an isolated event. In the 1990s, the U.S. government classified strong encryption as munitions under the Arms Export Control Act, creating a two-tier global landscape. U.S. firms either exported weakened security products or withdrew from international markets, leaving many countries reliant on inferior tools despite not being adversarial. The U.S. stance towards encryption didn’t change until open-source solutions emerged, and market dynamics rendered existing restrictions unfeasible. The European Union initially led the charge for regulatory liberalization, inadvertently benefiting African nations in the process.
The lifting of the June suspension has restored access uniformly, but if the AI Action Plan’s security assessment provisions solidify a sustained export control regime, a two-tier model—where full capabilities are available domestically while foreign access is limited—could soon become a reality.
Structural Challenges
The nearly three-week restriction on access to Anthropic’s AI models underscores a more substantial issue of dependency. Most countries, including those in Europe, rely heavily on U.S. cloud and AI infrastructure. However, African institutions typically face greater obstacles such as limited capital and insufficient domestic research capabilities. This broader vulnerability extends beyond AI, influencing debates on data localization in telecommunications and financial services. The African Union’s Malabo Treaty, although having taken effect in June 2023, was not crafted for scenarios in which a foreign government might suddenly disable access to critical technology.
Strategies for Mitigation
While no single solution exists, four strategies may help mitigate these dependencies:
1. **Localizing Infrastructure**: Mandating that frontier AI companies establish regional headquarters and host their capabilities locally can reduce latency issues and signal investment opportunities. While this approach may improve data access, it doesn’t completely insulate African institutions from export control directives, which typically apply regardless of where technology is housed.
2. **Diversifying Providers**: China’s Global AI Governance Initiative has emerged as a viable alternative for some African nations, as Chinese companies often provide AI solutions at lower costs than their Western counterparts. However, this dependency also carries risks, including potential governance issues and allegations of backdoor implementation.
3. **Investing in Local AI Development**: Building domestic AI capabilities through research partnerships and shared infrastructure can help alleviate long-term dependency. The AU’s Continental AI Strategy and the African Declaration on Artificial Intelligence prioritize infrastructure development, though these efforts may require significant time to yield results.
4. **Collective Bargaining**: African nations can enhance their negotiating power by forming alliances akin to the EU, allowing for a unified approach to discussions on regaining access to AI technologies. However, this strategy demands organizational cohesion that may come at the expense of speed.
The Urgency of Collective Action
These mitigation strategies are not mutually exclusive; however, they face an ordering challenge. Building local infrastructure and sovereign AI capabilities will take years, while decisions about access may require immediate action. Thus, collective bargaining emerges as the most urgent tool, not necessarily because it is the most powerful long-term solution but because it aligns with the immediacy of the problem.
African governments should take cues from the 1990s “cryptocurrency wars,” where collective EU action created an effective counterbalance. Currently, the bilateral dealings African nations pursue with the U.S. are producing fragmented outcomes that disadvantage individual countries. Without a unified front, each nation will approach negotiations from a place of vulnerability, accepting unfavorable terms.
The African Union’s stance on frontier AI access could serve as a critical starting point for necessary collective action.
Yasmin Abdillahi is a fellow at the Atlantic Council’s Africa Center.
