Explore the AI landscape.

Workers deserve a share of AI's gains and a real path forward when their jobs change.

Central question: What happens when companies no longer need as many human workers?

AI can write, summarize, code, and operate machines. Automating some tasks can shrink a workforce, increase output with the same staff, or create demand for different work; these outcomes should not be conflated.

AI's gains should reach the public, and a handful of suppliers should not decide who gets to compete.

Central question: Who owns the technology that could produce enormous amounts of wealth?

AI's economic gains depend on who owns chips, cloud infrastructure, models, applications, and rights to useful data. Concentration at one layer can give firms leverage over businesses at another, even when customers can choose among several chatbots.

Voters should know who is speaking and be able to trust information about how to vote.

Central question: Can democratic institutions function when convincing political content can be manufactured at scale?

Generative AI lowers the cost of producing political text, fake accounts, and realistic audio or video. It can also help citizens understand policy, translate official information, and challenge false claims; the political consequences depend on distribution, incentives, and audience response.

People should be able to move, communicate, and organize without being automatically tracked and profiled.

Central question: Will AI make mass surveillance significantly easier?

AI can connect faces, locations, purchases, messages, and behavioral patterns across large datasets. It changes the cost and reach of surveillance, even when the underlying cameras, commercial records, and collection powers already exist.

Schools should judge AI by what students learn, not how quickly it finishes their work.

AI can explain a problem, draft an essay, grade work, or simply supply the answer. The educational effect depends on which task the student still performs and how the tool is designed. A polished submission can conceal weak understanding, while a carefully designed tutor can support productive practice.

Powerful AI should have to demonstrate workable safeguards before it receives dangerous levels of access and autonomy.

Central question: How can humanity maintain control over increasingly capable AI?

Loss of control means a system persistently pursues behavior its operators cannot reliably constrain, correct, or stop. An existential catastrophe is a much stronger claim: an outcome that destroys humanity or permanently undermines its future; neither follows simply because a chatbot makes a mistake.

Set clear rules for serious risks, enforce them, and make compliance workable for smaller firms.

Central question: How much freedom should AI developers have?

AI governance is a choice about obligations, enforcement, and responsibility across different products and uses. A writing assistant, a hiring filter, and a highly autonomous frontier system do not create the same risks, so an effective framework should explain which rules apply to each.

A machine's speed should never erase human responsibility for using lethal force.

Military AI includes intelligence analysis, logistics, target recommendations, cyber operations, and weapons that select and engage targets after activation. These are different decisions: using software to sort imagery does not necessarily authorize it to kill someone. The central question is where human judgment must remain effective and accountable.

National security needs targeted controls, strong domestic research, and cooperation that prevents shared AI disasters.

US–China AI competition concerns semiconductors, computing infrastructure, research talent, models, commercial markets, and military applications. Leadership is not a single scoreboard: benchmark performance, manufacturing capacity, adoption, and security can point in different directions. Neither country operates independently of international suppliers and researchers.

Keep AI research and useful models accessible while treating irreversible releases of exceptional dangerous capabilities with care.

A model can be offered through a company-controlled service or released as downloadable weights that others can run and modify. Weights are the model's learned numerical parameters. Access to them does not necessarily include training data, training code, unrestricted licensing, or everything needed to reproduce the system.

Protect creators from piracy and exploitation while preserving lawful learning, fair use, and new creativity.

Generative AI raises separate questions about obtaining training material, using it to train, producing potentially infringing outputs, and protecting human work made with AI assistance. Copyright does not give someone ownership of every fact, idea, or artistic style. But copying protected expression, distributing pirated books, or reproducing substantial portions of works can raise serious legal issues.

AI companies should pay their infrastructure costs and disclose their resource use before communities bear the consequences.

AI infrastructure uses electricity for computing and cooling, water directly and through electricity production, and materials for chips and buildings. Training and everyday model use have different footprints. A single chatbot prompt is not a reliable stand-in for a data center's total environmental impact.

People deserve fair treatment, understandable reasons, and a real chance to challenge automated decisions.

Algorithms can help decide who gets a job, loan, apartment, insurance coverage, medical support, or criminal-justice intervention. Bias can enter through data, a poorly chosen objective, design choices, or deployment conditions. Removing a race or gender field does not prevent other variables from reproducing unequal treatment.

Emotional support should help people, not exploit their vulnerability or pretend to be proven mental-health care.

AI companions simulate conversation, attention, and sometimes intimacy; general chatbots can also become emotional supports even when not marketed that way. Some users feel heard, while others may rely on inaccurate advice or develop dependence. A system's supportive tone does not establish clinical competence or genuine understanding.

AI can help people act, but the authority to govern people must remain accountable to them.

AI can recommend a choice, make a choice, or carry it out. These are different transfers of authority: drafting a benefits letter is different from deciding eligibility, and suggesting a purchase is different from spending someone's money. The central question is who sets the goals, authorizes action, and remains answerable afterward.