Anthropic’s Mythos model negotiations expose systemic AI governance gaps amid national security and public interest tensions

Indigenous epistemologies view data as a sacred, communal resource governed by relational accountability, not corporate or state ownership. The Mythos model’s potential access by the US government contradicts these principles, framing data as a tool for control rather than a shared heritage. This clash reflects a deeper tension between extractive tech paradigms and Indigenous worldviews that prioritize reciprocity and consent. Without Indigenous oversight, AI governance risks replicating colonial patterns of resource extraction.

Historically, corporate-state AI collaborations have been justified under 'national security' or 'public good' narratives, from IBM’s collaboration with Nazi Germany to modern surveillance capitalism. The PRISM program revealed how tech giants enabled mass surveillance under legal cover, setting a precedent for unchecked data access. This case mirrors Cold War-era tech transfers to intelligence agencies, where short-term security gains obscured long-term democratic erosion. The lack of historical reflection in current discourse enables these patterns to repeat unchallenged.

Cross-culturally, AI governance models diverge sharply: the EU emphasizes rights-based regulation, China centers state-directed AI, and many Global South nations advocate for data sovereignty. Anthropic’s approach aligns with Silicon Valley’s libertarian-individualist ethos, which clashes with collective governance models in Africa and Latin America. The absence of Global South voices in this narrative reflects a broader failure to center non-Western perspectives in tech policy. This imbalance risks entrenching a unipolar AI governance regime.

Scientifically, the risks of AI model access by governments include bias amplification, adversarial manipulation, and unintended consequences in high-stakes decision-making. Studies show that even 'aligned' models can produce harmful outputs when deployed in unanticipated contexts, such as law enforcement or military applications. The lack of transparency in Anthropic’s training data and model architecture further complicates risk assessment. Without rigorous, independent audits, the Mythos model’s suitability for government use remains unproven.

Artistically and spiritually, AI access by governments evokes dystopian imagery of omniscient systems, from Orwell’s '1984' to the Hindu concept of 'Kali Yuga,' where knowledge becomes a tool of oppression. Creative resistance, such as algorithmic art that subverts surveillance, offers a counter-narrative to corporate-state AI dominance. Indigenous storytelling traditions warn against the hubris of 'playing god' with technology, emphasizing humility and interdependence. These perspectives highlight the ethical void in current AI governance debates.

Future scenarios include a bifurcated AI landscape: one where governments and corporations monopolize models for control, and another where open, community-owned AI fosters democratic innovation. The current trajectory risks a 'surveillance AI arms race,' where states and corporations compete to weaponize models, escalating global instability. Conversely, a rights-based approach could democratize AI, prioritizing public benefit over profit. The lack of proactive governance in this case suggests the former scenario is more likely without intervention.

Marginalized communities, including racial minorities, low-income groups, and Global South nations, are systematically excluded from AI governance decisions despite bearing disproportionate risks. Black and Indigenous activists have long warned about predictive policing algorithms reinforcing systemic racism, yet their expertise is sidelined in corporate-state negotiations. The Mythos model’s potential use in immigration enforcement or military applications directly threatens these groups, yet their voices are absent from the discourse. True accountability requires centering their lived experiences and demands.