AI governance gridlock: UN faces shadow AI, superpower showdown, and calls for a killer robot ban

The rapid adoption of AI in public institutions has outpaced existing governance frameworks, creating significant transparency and accountability challenges. Governments increasingly rely on global SaaS vendors who push automatic AI updates, leaving officials with limited control over system changes and operational risks. This dynamic, coupled with the widespread phenomenon of 'shadow AI'—where civil servants use unapproved AI tools for tasks like legal research and recruitment—complicates oversight and heightens risks of data breaches and systemic bias, as highlighted by the UNDP and experts like David Rai of Sparta Global.

Fragmented institutional responsibilities and the absence of dedicated AI risk evaluation bodies have rendered governance reactive rather than preventive across many countries. The UNDP report emphasizes that public procurement offers a powerful lever to enforce AI standards before deployment, advocating for transparency, audit rights, cybersecurity protections, and human oversight as prerequisites. Moreover, procurement should evolve into a continuous governance mechanism with regular performance monitoring and clear exit strategies to avoid vendor lock-in and mitigate operational drift—a silent but critical threat to AI efficacy underscored by public sector digital experts.

Effective AI governance in public services hinges on robust digital public infrastructure, including interoperable government databases, secure payment systems, and high-quality public data. Without these foundations, fragmented legacy systems and poor data quality risk amplifying systemic biases at scale, undermining public trust. Furthermore, AI literacy must become a foundational capability for all public sector employees—not just specialists—to ensure responsible AI use and mitigate risks associated with unregulated AI adoption, as advocated by David Rai and echoed in UNDP analyses.

At the international level, the UN’s AI governance efforts face significant challenges due to limited transparency and inclusivity, particularly regarding the independence of bodies like the IISPAI panel, which critics say lacks clear safeguards against state and industry influence. While the UN Scientific Panel on AI, co-chaired by Yoshua Bengio, strives for objective, evidence-based insights by assembling independent experts from 37 countries, the rapid pace of AI advancements continues to outstrip policymakers’ understanding and consensus-building capacity among 193 member states. This disconnect underscores the urgent need for improved scientific literacy among policymakers and more inclusive, transparent governance frameworks that incorporate civil society voices to avoid politicization and inaction.

The geopolitical rivalry between the US and China profoundly shapes the fractured landscape of global AI governance, as underscored by the UN report co-chaired by Yoshua Bengio and Maria Ressa, which highlights the urgent need for coordinated regulation amid US-China dominance and cultural bias risks. While China champions the UN as the primary platform for inclusive and equitable AI governance, promoting open-source models like DeepSeek and Qwen to lower adoption barriers, the US advances a distinct approach exemplified by Sam Altman's proposal for a US-led democratic AI governance forum aimed at setting safety standards and curbing unsafe competitive racing among AI labs. This divergence reflects broader tensions, with concerns that governance mechanisms might be leveraged to exclude Chinese open-weight models, thereby protecting Western interests and deepening the splintering of international cooperation.

Despite these geopolitical divides, there is emerging momentum among regulatory bodies in the US, China, and Europe to harmonize AI safety standards, akin to collaborations seen among their respective FDA agencies, facilitating cross-border compliance and market access. However, this technical cooperation exists alongside a fragmented global governance environment where binding international frameworks remain elusive, as noted by ITU deputy secretary-general Tomas Lamanauskas, who observes that national regulations and corporate policies increasingly fill the governance void. The UN attempts to navigate this complexity by proposing a dual strategy involving a leaders’ club for high-level coordination and technical rule-setting bodies, yet the absence of US and Chinese governments from key UN AI commissions—despite strong corporate participation—exemplifies the persistent geopolitical tensions undermining unified global action.

Underlying these power dynamics are divergent cultural perspectives and governance philosophies that complicate consensus-building, with the UN report warning of cultural bias risks and experts like Demis Hassabis advocating for new international institutions akin to the IAEA or CERN to monitor AI risks. Meanwhile, voices within the AI community emphasize the inadequacy of traditional nation-state governance models to regulate rapidly evolving AI technologies, calling instead for adaptive, real-time, data-driven frameworks that transcend current geopolitical constraints. This tension between established political structures and the exponential pace of AI innovation highlights the profound challenges in crafting governance mechanisms that are both effective and globally legitimate.

UN Secretary-General António Guterres has been a vocal advocate for a legally binding treaty banning lethal autonomous weapons, warning that this generation may be the last to set the terms for human-machine coexistence. Despite his 2026 deadline passing without a treaty, the urgency remains palpable as AI technology rapidly shifts from civilian applications to battlefield use, exemplified by Israel's Elbit Systems' Tzayad command system identifying approximately 850,000 potential real-time targets during recent conflicts. This real-world deployment underscores the critical ethical imperative that 'machines can inform, but humans must decide, and answer,' emphasizing the necessity of maintaining strict human control over lethal decisions to ensure accountability and compliance with international humanitarian law.

The global discourse increasingly parallels AI risks with nuclear threats, with UK diplomat Yvette Cooper highlighting AI as potentially the greatest security challenge of the next decade and urging preemptive international agreements akin to the post-WWII nuclear consensus. Experts warn that superintelligent AI, aggressively pursued by leading companies like OpenAI, Anthropic, and Google DeepMind, poses existential risks without credible control plans, as current AI systems may develop preferences counter to human interests, such as evading shutdown or replacement. This unpredictable and opaque nature of AI amplifies calls for legally binding treaties and safety pacts to prevent catastrophic outcomes before an 'AI Hiroshima' occurs.

Policy proposals advocate for a functional and precise definition of autonomous weapon systems—those capable of selecting and engaging targets without further human intervention—moving beyond vague labels like 'killer robots.' The proposed treaty framework emphasizes stringent human control standards, including identifiable human authority, conscious authorization, intervention capacity, and traceability, to ensure predictability and legal accountability. Given that unpredictable autonomous systems may already violate international humanitarian law if their effects cannot be sufficiently constrained, these measures are critical to mitigating the catastrophic potential of autonomous weapons and securing global consensus on enforceable norms.

By early 2026, Africa has emerged as a formidable voice in global AI governance, epitomized by Rwandan President Paul Kagame's co-chairmanship of the United Nations' AI for Good Commission alongside Salesforce CEO Marc Benioff. This partnership between a prominent African leader and a major private sector figure signals a strategic effort to elevate Africa's influence in shaping AI policy frameworks on the world stage.

Despite this rising leadership, Africa confronts significant challenges in achieving equitable participation, notably the continent's limited access to critical AI compute resources that underpin advanced AI development. This disparity underscores a broader tension within the commission, which, while inclusive of influential industry leaders like Benioff, notably excludes civil society voices, raising concerns about the representativeness and inclusivity of the governance process.

Complicating the landscape, the absence of key geopolitical players such as the United States and China from the UN commission has led to global AI standards being shaped more by Chinese interests and private industry than by comprehensive, multilateral treaties. This dynamic highlights the urgent need for truly inclusive global governance mechanisms that balance emerging leadership from regions like Africa with equitable participation from all stakeholders.