AI’s life-or-death inconsistency shows why we need decentralization

A recent RAND Corporation study published in Psychiatric Services revealed a chilling truth about our most trusted AI systems: ChatGPT, Gemini, and Claude respond dangerously inconsistently to suicide-related queries. When someone in crisis asks for help, the response depends entirely on which corporate chatbot they happen to use.

This isn’t a technical bug that can be patched in the next software update. It’s a serious failure of trust that exposes the fundamental flaws in how we build AI systems. When the stakes are literally life and death, inconsistency becomes unacceptable.

The problem runs deeper than poor programming. It’s a symptom of a broken, centralized development model that concentrates power over critical decisions in the hands of a few Silicon Valley companies.

The black box problem

The safety filters and ethical guidelines governing these AI systems remain proprietary secrets. We have no transparency into how they make critical decisions, what data shapes their responses, or who determines their ethical frameworks.

This opacity creates dangerous unpredictability. Gemini might refuse to answer even low-risk mental health questions out of excessive caution, while ChatGPT could inadvertently provide harmful information due to different training approaches. Legal teams and PR risk assessments more often govern the responses than by unified ethical principles.

A single company cannot design a one-size-fits-all solution for global mental health crises. The monolithic approach lacks the cultural context, nuance, and agility required for such sensitive applications. Silicon Valley executives making decisions in boardrooms cannot possibly understand the mental health needs of communities across different cultures, economic conditions, and social contexts.

Community auditing beats corporate secrecy

The solution requires abandoning the closed, centralized model entirely. Critical AI safety protocols should be built like public utilities — developed openly and auditable by global communities of researchers, psychologists, and ethicists.

Open-source development enables distributed networks of experts to identify inconsistencies and biases that corporate teams miss or ignore. When safety protocols are transparent, improvements happen through collaborative expertise rather than corporate NDAs. This creates competitive pressure toward better safety outcomes rather than better legal protection.

Community oversight also ensures that cultural and contextual factors are properly addressed. Mental health professionals from different backgrounds can contribute specialized knowledge that no single organization possesses.

Infrastructure determines possibilities

Building robust, transparent AI systems requires neutral infrastructure that operates independently of corporate control. The same centralized cloud platforms that power current AI giants cannot support genuinely decentralized alternatives.

Decentralized compute networks, like those we are already seeing with io.net, provide the computational resources necessary for communities to build and operate AI models without dependence on Amazon, Google, or Microsoft infrastructure. This technical independence enables genuine governance independence.

Community governance through decentralized autonomous organizations could establish response protocols based on collective expertise rather than corporate liability concerns. Mental health professionals, ethicists, and community advocates could collaboratively determine how AI systems should handle crisis situations.

Beyond chatbots

The suicide response failure represents a broader crisis in AI development. If we cannot trust these systems with our most vulnerable moments, how can we trust them with financial decisions, health data, or democratic processes?

Centralized AI development creates single points of failure and control that threaten society beyond individual interactions. When a few companies determine how AI systems behave, they effectively control the information and guidance that billions of people receive.

The concentration of AI power also limits innovation and adaptation. Decentralization unlocks greater diversity, resilience, and innovation — allowing developers worldwide to contribute new ideas and local solutions. Centralized systems optimize for broad market appeal and legal safety rather than specialized effectiveness. Decentralized alternatives could develop targeted solutions for specific communities and use cases.

The moral infrastructure challenge

We must shift from comparing corporate offerings to building trustworthy systems through transparent, community-driven development. Technical capability alone is insufficient when ethical frameworks remain hidden from public scrutiny.

Investing in decentralized AI infrastructure represents a moral imperative as much as a technological challenge. The underlying systems that enable AI development determine whether these powerful tools serve public benefit or corporate interests.

Developers, researchers, and policymakers should prioritize openness and decentralization not for efficiency gains but for accountability and trust. The next generation of AI systems requires governance models that match their societal importance.

The stakes are clear

We’re past the point where it’s enough to compare corporate chatbots or hope a “safer” model will come along next year. When someone is in crisis, their well-being shouldn’t depend on which tech giant built the system they turned to for help.

Consistency and compassion aren’t corporate features; they’re public expectations. These systems need to be transparent and built with the kind of community oversight that you get when real experts, advocates, and everyday people can see the rules and shape the outcomes. Let’s be real: the current top-down, secretive approach hasn’t passed its most important test. For all the talk of trust, millions are left in the dark (literally and figuratively) about how these responses are set.

But change isn’t just possible, it’s already happening. We’ve seen, through efforts like those at io.net and in open-source AI communities, that governing these tools collaboratively isn’t some pipe dream. It’s how we move forward, together.

This is about more than technology. It’s about whether these systems serve the public good or private interest. We have a choice: keep the guardrails locked in boardrooms, or finally open them up for genuine, collective stewardship. That’s the only future where AI truly earns public trust and the only one worth building.