The Accountability Playbook: Assign Responsibility Before AI Scales

A three-way split of responsibility

David Solomon, chairman and chief executive of Goldman Sachs, told the firm's Exchanges podcast that the current AI debate is too binary, framed as a choice between innovation without guardrails and heavy-handed regulation. Solomon argues that responsibility for AI control should be distributed among governments, which set rules, model developers, which bear responsibility for model safety, and the businesses deploying the technology, which remain accountable for how it is used. Goldman Sachs says it has used AI-type models in trading platforms for nearly 20 years with human accountability attached to every platform, and Solomon argues financial institutions can offer risk management and audit practices to help other businesses deploy AI safely. m:…3249

Solomon also addressed jobs directly, arguing that AI will disrupt and automate jobs but that the pace of change and the creation of new businesses will act as counterbalances, and that market volatility in the AI sector reflects uncertainty about the pace of development and ultimate demand for compute infrastructure. Goldman Sachs is committing 500 million dollars to launch the Small Business AI Institute, and Solomon noted that small businesses account for 99% of all companies in the United States and drive 60 to 65% of annual job creation. The institute is therefore a bet that the accountability question will be decided in small firms as much as in large ones, since that is where most jobs are created. m:…3249

Healthcare: use AI to expose the contract, not just treat the patient

Mark Cuban, founder of Cost Plus Drugs, made a parallel argument from healthcare on the a16z podcast, saying the system lacks competitive dynamics because it lacks price discovery. Cuban argues employers can significantly reduce healthcare costs by running benefits contracts through AI to identify unfavorable terms, and he claims that if 67% of employers audited their contracts with AI, the current vertically integrated insurance and pharmacy benefit system would collapse. That collapse claim is Cuban's argument, not an established outcome, and it should be read as a provocation about incentives rather than a forecast. Cuban also proposes a policy solution: mandating public disclosure of all contracts involving taxpayer money. m:…5986

Cuban's own company illustrates the transparency play. Cost Plus Drugs has grown from 111 medications at launch to approximately 4,000 SKUs, pays independent pharmacies a 12 dollar fill fee to support local pharmacy infrastructure, and is moving into modular manufacturing pods that cost approximately 20 million dollars each. Cuban cites UnitedHealth managing over 3,000 subsidiaries and conducting 150 billion dollars in annual intercompany transfers as evidence of the opacity he is arguing against. His view on clinical AI is more restrained: AI will not replace doctors, he argues, but doctors who know how to use AI will replace those who do not, which puts the accountability question back on training and workflow rather than on the model alone. m:…5986

The public-sector version: boring workflows, honest effort

Jake Burns, executive-in-residence at AWS, gave the public-sector version of the same lesson on the Tyler Tech Podcast, defining agentic AI as generative models given access to tools that let them take action in the world, such as database writes or email sending. Burns argues that AI systems often create more work for users rather than reducing it, because higher effort is required to achieve productivity gains, and that organizations fail to move from pilot to production because they lack a bridge phase that replicates production-like messiness and guardrails. Workforce buy-in is the most critical factor for successful adoption, Burns says, and the most successful implementations in large enterprises are often simple, multi-step workflow automations rather than fully autonomous systems. m:…5771

The three accounts converge on a usable playbook. Assign responsibility across government, developer and deployer before a system scales, as Solomon proposes. Use AI first where it creates auditability, such as contract review, as Cuban proposes, rather than where it merely promises savings. And ship small, boring automations with a bridge phase and workforce buy-in, as Burns proposes, instead of leaping from prototype to autonomous system. None of the three promises that this playbook removes risk, but together they describe how accountability moves from a debate topic to an operating discipline. m:…3249 m:…5986 m:…5771