Anthropic’s Mythos AI: How corporate secrecy and unchecked tech development fuel systemic cybersecurity risks

Indigenous cybersecurity frameworks emphasize relational accountability, where technology is embedded in ecological and social systems rather than treated as a standalone commodity. The Māori principle of *whanaungatanga* (relationship-building) could reframe AI risks as collective challenges requiring kinship-based solutions, such as co-designed governance models. Current approaches, however, ignore these frameworks, treating cybersecurity as a purely technical problem solvable through proprietary tools. This omission reflects a deeper colonial logic that devalues non-Western knowledge in critical infrastructure.

The unauthorized access to Mythos echoes historical patterns of tech disruptions, from the 1988 Morris Worm to the 2017 NotPetya attack, which exploited systemic vulnerabilities in global digital infrastructure. These incidents reveal a recurring cycle: corporations prioritize speed and profit over security, regulators lag behind, and the public bears the costs. The tech industry’s 'move fast and break things' ethos has repeatedly led to crises that are only addressed retroactively, often through costly bailouts or reactive legislation. Mythos fits this pattern, with Anthropic’s secrecy mirroring past failures like the 2013 NSA leaks, where internal controls were bypassed in the name of innovation.

Cross-culturally, cybersecurity is often framed as a public good rather than a private asset, with solutions emerging from communal practices like India’s *Cyber Gram Yojana* or Brazil’s *Rede de Proteção Digital*. In Japan, the concept of *wa* (harmony) informs collaborative cybersecurity strategies, contrasting with the adversarial models dominant in Western cyber warfare. These approaches highlight how different cultures conceptualize risk and trust, suggesting that global solutions require pluralistic governance rather than Silicon Valley’s one-size-fits-all models. The current crisis underscores the need for these diverse perspectives to inform international AI standards.

From a scientific standpoint, the risks of unauthorized access to Mythos are well-documented in cybersecurity literature, particularly regarding adversarial attacks on large language models (LLMs). Studies show that even restricted models can be jailbroken or fine-tuned for malicious purposes, as seen in research from Carnegie Mellon and Stanford in 2024. However, Anthropic’s opacity limits independent verification of its security claims, raising questions about the reproducibility of its risk assessments. The incident also reflects a broader scientific failure: the lack of standardized benchmarks for AI safety, which enables companies to self-regulate without external scrutiny.

Artistically and spiritually, the Mythos AI incident can be read as a modern myth of hubris, where human ambition outpaces ethical and ecological limits. Indigenous stories like the Greek tale of Prometheus warn of the dangers of stealing fire (knowledge) without responsibility, a metaphor that resonates with today’s AI landscape. Creatively, artists like Refik Anadol have explored AI as a collaborative tool, challenging the dominant narrative of control and competition. These perspectives suggest that reimagining AI requires not just technical fixes but a cultural shift toward humility and reciprocity in technological development.

Future modelling scenarios suggest that without structural changes, unauthorized access to advanced AI models will become the norm, not the exception, as global actors—states, corporations, and non-state groups—compete for advantage. A 2025 RAND Corporation report warns that by 2030, AI-driven cyberattacks could disrupt critical infrastructure at scale, with cascading effects on food, energy, and healthcare systems. However, alternative futures are possible: if governance frameworks like the EU AI Act are strengthened and global south voices are centered, AI could be deployed in ways that prioritize resilience over profit. The current trajectory risks locking in a dystopian scenario where cybersecurity is a luxury, not a right.

Marginalized voices—such as gig workers labeling data for AI systems, communities affected by algorithmic bias, and Global South cybersecurity researchers—are systematically excluded from narratives about AI risks. These groups bear the brunt of AI’s failures, from exploitative labor practices in data annotation to the disproportionate impacts of AI-driven surveillance in marginalized communities. For example, workers in Kenya and the Philippines, who train AI models for Silicon Valley firms, often face precarious conditions with no recourse when systems they helped build are weaponized. Their exclusion from the Mythos debate reflects a broader pattern of techno-colonialism, where risks are globalized but accountability remains local.