An Arizona judgment exposes the difference between giving people a voice and claiming their authority. The economics of synthetic speech will turn on who can consent, who gets paid and who must prove what is real.

On 30 September, an Arizona appeals court upheld a manslaughter conviction but ordered a new sentencing. The sentencing judge had relied on an AI video depicting the deceased victim speaking. The ruling concerns a particular use of unreliable material, not every use of AI in a courtroom. Its wider significance is unsettling: making someone appear present can give other people's words the authority of their identity.

Society should welcome tools that help people express themselves. But the right to generate a convincing likeness should not become the right to speak for its subject. That distinction matters wherever a familiar voice can influence grief, money, reputation or freedom.

The video had disclosed its AI origins. The court nevertheless found that it presented imagined thoughts as the victim's own; it distinguished this reconstruction from genuine recordings embedded within it. Disclosure therefore did not resolve the problem in this case. Knowing how an image was produced did not establish whose judgment it expressed.

The family's wish to make a loved one present deserves respect. There is a serious case for allowing expressive technologies to help people communicate what conventional language cannot. A blanket rejection could also obstruct valuable uses: the US Federal Trade Commission has identified potential assistance for people who have lost their voices. The ethical question is who controls the resulting speech and what others are entitled to infer from it.

Consent has several sides. The person whose identity is reproduced, the person commissioning it and the person encountering it may want different things. Cambridge researchers' 2024 study of digital afterlife services recommends consent from both the people represented and those interacting with their simulations. This is ethical analysis using imagined scenarios, not proof that such products either heal or damage bereavement.

The commercial tension is plain. A service might earn recurring revenue from continued interaction with a simulated relative. Ending that subscription could then feel unlike cancelling ordinary software. That is a foreseeable business-model risk, not an allegation about the family in Arizona. A humane service would make limits and eventual closure part of its design.

Work presents a related bargain. Synthetic speech can reduce the need for repeated recording sessions; it can also create demand for authorized voices, editing and review. The FTC identifies appropriation of voice artists' identities as a livelihood risk. Which outcome prevails depends partly on contracts: whether performers approve particular uses, share in reuse revenue and retain a meaningful ability to refuse. Cheaper production does not automatically distribute the savings fairly.

There is another bill: verification. A convincing clip may take little effort to circulate, while challenging it can require expertise, records and time. If platforms collect the revenue while individuals must finance every dispute, cheaper communication can make trustworthy communication less equal. Wealthier people and institutions could purchase assurance that others cannot afford.

India has a substantial stake in getting this bargain right. In February, the government announced an RBI–BHASHINI agreement aimed at multilingual banking access, including voice interfaces. This is an announced programme, not evidence of achieved inclusion, and language assistance need not imitate an identifiable person. It illustrates why understandable speech can be valuable without borrowing someone's identity.

For a customer navigating banking in a familiar language, access and trust must advance together. Institutions should provide a way to confirm who is speaking and challenge consequential instructions. An English disclaimer or an app-only verification step would be a poor safeguard for someone facing language, literacy or connectivity barriers. Human assistance through telephone and branch channels should remain available.

Three futures are plausible. In a constructive one, people gain affordable expression, performers license their work on fair terms, and institutions absorb verification costs. In an uneven one, synthetic services spread while reliable channels and effective remedies become privileges. In a harmful one, simulated intimacy becomes a sales technique and uncertainty helps both impersonators and people dismissing authentic evidence.

Technology alone cannot select between them. The FTC's 2024 review of voice-cloning safeguards describes prevention, detection and later evaluation, while warning that watermarks can be altered or removed. Institutions must decide what authority a recording is allowed to carry, even when everyone knows it is synthetic.

Whose permission makes this product legitimate? Does the person represented share in its value? Who pays when attribution is disputed? Can users withdraw, and can affected people reach a human decision-maker?

Build consent, compensation and dispute handling into the price. Preserve original recordings and identify who authored generated words. A voice should help a person exercise agency; its convincing imitation should never be sufficient authority to take that agency away.