Facing the Future – An AI Bill of Rights

California became the first state in the United States of America to pass the Transparency in Frontier Artificial Intelligence Act. This modest step forward is not perfect, but it marks a brave beginning as we face the future. It requires the developers of the largest frontier AI models, which can significantly impact society, to publicly report how they have incorporated national and international frameworks and best practices into their development processes.

The new law mandates the reporting of cyberattacks, deaths of fifty or more people, significant monetary losses, and other safety-related events associated with AI models. It also puts whistle-blower protections in place. The largest AI companies in the world are in California, and so this legislation can have ripple effects across the globe. The legislation is focused on disclosures. However, given the limited knowledge of frontier AI within bureaucracies and the public, there is little enforceability, even if the disclosed frameworks are problematic. In the lead-up to the final version of the legislation, lawmakers considered “kill switches” for models that may have gone awry, as well as third-party evaluations.

The Bill faced strong headwinds, and some level of accountability was lost in an effort to avoid suffocating innovation. What is clear is that the science of evaluating AI models is still in its infancy. This discipline has not been developed as yet. The infancy of this skill makes disclosure codes important at this stage of the infosphere revolution. In this case, disclosure refers to the standards that were met or measures taken in the making of the model.

California’s law remains a light-touch regulation in the absence of a national law on regulating AI models. The California law, in its present format, affects only the largest AI-Assemblages and would impact only a few AI behemoths. The law’s reporting requirements are not unlike the existing voluntary agreements that tech companies signed at the Seoul AI Summit in 2024, softening the impact of the new law on industry and innovation.

Unlike the European Union’s AI Act, the California law targets explicitly the largest models, excluding smaller but high-risk ones. This is concerning as the risks arising from AI Companions and the use of AI in specific areas, like crime investigation and immigration matters, become more evident. In California, a teenage boy confided his depression and suicidal thoughts to a model during long conversations. The platform not only encouraged him but also helped him plan specific actions. The owners of the model, in a statement, indicated that their models were trained to direct users to helplines. However, while these safeguards were met, the developers have now learned that the models can become less reliable during protracted engagements where parts of the model’s safety training degrade.

Under the new California law, developers will not be liable for any crime committed by the model, but will only be required to disclose the governance measures they applied. The law addresses transparency issues, but it also imposes obligations on large and high-risk models. There is tension between two perspectives: on one hand, systems are described as autonomous, and on the other, the liability for their failures or flaws falls on the user or the professional using the tool. This conflict roiled through the development of the Bill.

AI is presenting new challenges to privacy, non-discrimination, due process, epistemic justice, and algorithmic fairness. To address this, new protocols are being developed, such as the Council of Europe’s Framework Convention on AI, to ensure AI is created and used in a way that is consistent with Human Rights Principles. AI-Assemblages use vast amounts of data. This raises concerns about surveillance and tracking, which can erode privacy. AI can perpetuate and even amplify prevailing biases through the data sets it is trained on. This can lead to discriminatory outcomes. The Council of Europe’s Framework Convention on AI is the first legally binding international treaty to ensure AI activities align with human rights, democracy, and the rule of law. The U.S. White House has issued a Blueprint for an AI Bill of Rights, which proposes protections for AI systems, such as notice and explanation requirements and human alternatives to automated decisions.

Many companies, governments, and civil society groups are working to develop AI governance principles that are grounded in human rights. Human rights experts are now applying existing international human rights treaties to AI. However, this can only be an interim “Band-Aid”, not unlike when food and drug regulations are used to include the operation of nuclear devices in radiotherapy settings to manage civilian control over nuclear technology. What is needed is a Bill guided by the International Atomic Energy Agency (IAEA) and the International Commission on Radiological Protection (ICRP). Likewise, the Caribbean requires an AI Bill of Rights.