Systemic risks of AI agent proliferation: How corporate governance fails to secure non-human identities amid agentic AI expansion

Indigenous knowledge systems universally treat non-human entities as subjects with agency, not objects to be controlled, which challenges the premise of 'agent-first' governance as inherently extractive. Traditional governance models, such as the Māori *kaitiakitanga* (guardianship), prioritize long-term ecological and social balance over short-term efficiency, offering a corrective to the profit-driven deployment of AI agents. The absence of these perspectives in the narrative reflects a broader erasure of relational ontologies in tech governance, where AI agents are framed as neutral tools rather than participants in complex socio-ecological systems.

Historical parallels abound in the automation of governance, from the Luddite rebellions against mechanized textile looms to the 2008 financial crisis, where algorithmic trading systems amplified systemic risk. The Boeing 737 MAX disasters demonstrated how corporate incentives to cut costs and accelerate deployment led to catastrophic failures in automated systems, mirroring today’s rush to deploy AI agents without adequate oversight. The concept of 'non-human identities' echoes earlier debates about corporate personhood, where legal fictions granted rights to entities (e.g., corporations) at the expense of human stakeholders—a precedent now repeating with AI agents.

Cross-cultural comparisons reveal that non-Western societies often embed AI governance within broader ethical frameworks, such as Ubuntu philosophy in Southern Africa, which emphasizes communal well-being over individual efficiency. In contrast, Western corporate models treat AI agents as isolated, optimizable units, ignoring their embeddedness in social and ecological networks. The EU’s risk-based approach to AI regulation, while imperfect, contrasts sharply with the US’s laissez-faire model, highlighting how cultural values shape governance priorities.

Scientific literature on AI security consistently shows that agentic systems introduce novel attack vectors, including prompt injection, model theft, and supply-chain vulnerabilities, which are poorly addressed by existing frameworks like NIST’s AI Risk Management Framework. Research on non-human identities (NHIs) in cybersecurity is nascent but growing, with studies indicating that NHIs are 3-5x more likely to be compromised than human identities due to weaker authentication protocols. The lack of standardized benchmarks for agent security further compounds the problem, as enterprises deploy agents without rigorous stress-testing.

Artistic and spiritual traditions often depict non-human entities as tricksters or guardians, reflecting the dual-edged nature of AI agents—capable of both liberation and exploitation. The concept of *anima* in Jungian psychology warns of the dangers of projecting human-like agency onto non-human systems, a phenomenon increasingly observed in AI agent deployments. Indigenous storytelling, such as the Māori *taniwha* (water spirits), frames non-human entities as entities to be negotiated with, not controlled—a metaphor for the need to rethink AI governance as a relational, not transactional, process.

Future scenarios modeling the proliferation of AI agents predict a 200-400% increase in cyber-physical system failures by 2030 if governance frameworks remain reactive rather than proactive. Agentic AI could enable autonomous cyber warfare, where state and non-state actors weaponize AI agents to disrupt critical infrastructure, necessitating international treaties akin to the Chemical Weapons Convention. The rise of 'agent swarms'—decentralized, self-organizing AI collectives—poses existential risks to global supply chains, requiring scenario planning that includes worst-case disruptions.

Marginalised communities, including gig workers, low-wage employees, and Global South populations, bear disproportionate risks from AI agent proliferation, as their labor is increasingly automated or surveilled by these systems. The framing ignores how AI agents in customer service or HR replicate and amplify existing biases, disproportionately affecting women and people of color. Indigenous data sovereignty movements, such as the *First Nations Information Governance Centre* in Canada, highlight how unchecked AI agent deployment threatens cultural survival by commodifying traditional knowledge without consent.