Promises Versus Statutes: Who Will Rule Artificial Intelligence at Work?

Washington chose voluntary pledges, California chose binding law. India must decide how to protect workers from automated decisions.

Two things happened in the same week, one on each side of America, and together they tell a bigger story about how the world is trying, and struggling, to govern artificial intelligence.

On Tuesday, at the White House, President Donald Trump said top technology leaders had agreed to a voluntary accord to police themselves. the executives came to the table after several high-profile mishaps and warnings from some of their own colleagues about the risks of the technology. A day later, California Governor Gavin Newsom signed a set of laws meant to protect workers from AI-related risks, including the threat of lost jobs and the spread of workplace surveillance. Earlier in the month, he had signed a law requiring operators of AI chatbots to assess risks before releasing them, and an order requiring the state to consult experts on oversight.

One approach asks companies to promise. The other tells them what they must do. Which one wins will matter for workers everywhere, including in India, where millions of jobs sit in industries that automation can reach.

What a promise can do

A voluntary pledge is not worthless. It can set expectations quickly. It can bring competitors to the same table. It can also draw on the knowledge of people who understand the technology far better than most lawmakers do. A law written in a hurry can be badly drafted and may even make things worse.

There is a practical reason many governments start here. AI is changing fast. A rule that fits today's tools could be out of date next year. Voluntary codes are easier to update.

But promises have limits, and they are well known.

• They cannot be enforced. If a company breaks the pledge, there may be no penalty beyond bad press.

• They are written by those they bind. The people who gain from fast deployment decide how fast to go.

• They rarely cover workers. Safety pledges often focus on dramatic risks such as misuse or loss of control. Everyday harms, like being scored by an algorithm or replaced without notice, get less attention.

• They can reduce pressure for law. The appearance of action can make it harder to pass real rules.

The trigger for the White House pledge also tells a story. It reportedly followed mishaps and warnings, which suggests that voluntary measures often arrive after harm, not before.

What a statute can do

California's approach is different. A law says what must happen, who must do it and what the penalty is. It applies to every company, not only those that sign up. It gives workers a way to complain and a regulator a way to act.

Reports say the new laws deal with two kinds of risk that are easy to overlook. The first is job loss. When a company plans to replace workers with automated systems, a law can require notice, consultation or support. The second is surveillance. Tools that track keystrokes, measure speed, listen to calls or score behaviour can make work harder, and many workers do not know what is being measured or how it is used. A law can require disclosure and set limits.

The earlier law on chatbots adds a third layer. Companies must assess the risks before they release the product, rather than finding out after people are hurt.

There are fair criticisms, too. A patchwork of state laws can make life hard for companies that operate nationwide, since they must meet different rules in different places. Companies argue that heavy rules may push investment elsewhere. Some experts say job-loss rules are difficult to enforce, because it is hard to prove that a particular layoff was caused by AI and not by other factors. These are serious points, and the results of the California laws will take time to understand.

A fight over words

The two approaches came with a small, revealing argument over language. Mr Newsom signed an order requiring state agencies to keep calling the technology "artificial intelligence". This was a reply to a recent order by Mr Trump that told US diplomats to use the term "super intelligence".

It might sound petty, but words shape policy. "Artificial intelligence" suggests a tool that people use and can control. "Super intelligence" suggests something beyond human control, which can lead to either panic or a sense that nothing can be done. How a government talks about the technology affects what it thinks it is allowed to regulate.

Mr Newsom also said the country needs to do a lot more in the absence of federal leadership, and he did not rule out calling a special session of the legislature. His term ends in January, and midterm elections are weeks away, so the politics are never far from the policy.

The money behind the debate

The scale of money in this industry explains why the stakes are high. OpenAI is reportedly looking to raise about $30 billion at a valuation near $1.4 trillion, and its chief executive has said it will not go public until its models are safe. A separate frontier lab has filed an IPO prospectus that shows deep reliance on large technology partners. Broadcom is reportedly gathering about $60 billion to fund chips for one lab.

When this much money is at stake, the pressure to release products fast is intense. That is exactly the setting in which voluntary promises are tested hardest. It is also the setting in which binding rules give workers and the public something firmer to rely on.

When the money is this large, a promise is only as strong as the incentive to keep it.

Why India should pay attention

India is not just a user of AI. It is one of the world's largest workforces in services that AI can touch. Information technology, business process outsourcing, customer support, back-office work and parts of banking and insurance employ millions of people. Many of these jobs involve tasks, such as writing, coding, data handling and call-handling, that AI tools already perform to some degree.

At the same time, India is eager to adopt AI. The government hosted the India AI Impact Summit in February, with a focus on use for citizens and the Global South. Public bodies are building AI into services. The Union Cabinet recently cleared an AI-enabled traffic management system for Delhi at a cost of ₹1,789.52 crore. The Finance Ministry has said that limited participation in global AI developments weighs on India's appeal to investors.

So India wants to move faster, and it also has more workers exposed to the change. Yet, as far as we are aware, India relies mainly on general data protection and information technology laws, such as the Digital Personal Data Protection Act of 2023, and not on a dedicated law for AI in the workplace. Summit declarations are not enforceable rules.

Questions workers might reasonably ask

Think about an ordinary employee in an Indian city. She works in a call centre or a bank's back office. Her employer introduces a system that scores her calls, measures her pauses and flags her for review if her numbers dip. Or the employer announces that a chatbot will now handle a large share of the queries she used to answer.

She might ask a few simple questions.

• Will I be told when a machine is measuring or judging my work?

• On what basis is the system scoring me, and can I challenge it?

• If my role is cut, will I get notice, training or help to move?

• Who is responsible if the system makes a mistake about me?

Today, in many workplaces, nobody is required to answer any of these. That is the gap that California is trying to fill, and the gap that India will need to consider.

What a sensible Indian approach could include

India does not need to copy California. It has different labour laws, a different economy and different institutions. But it can learn from the choice between promises and statutes. A balanced approach might include:

1. Disclosure. Employers should tell workers when automated systems monitor or evaluate them, and on what broad criteria.

2. Impact assessments for large deployments. Big employers and public bodies should assess how a system affects workers and citizens before using it widely.

3. A right to human review. Workers should be able to ask for a person to look again at a significant decision, such as dismissal or demotion, made with the help of AI.

4. Transition support. Where jobs are cut because of automation, funds for retraining and placement should be available, ideally from the companies gaining the savings.

5. Public data. Regular publication of how AI is changing jobs, by sector, so that policy can be based on facts.

6. Flexible rules. Laws should set principles and leave room for regulators to update details as technology changes, which addresses the main objection to hard rules.

This list does not ban or slow AI. It sets a floor of fairness while allowing adoption.

A fair view of both sides

The voluntary approach should not be dismissed. Industry knowledge is valuable, and a pledge can be the first step toward rules. The risk is when it becomes the last step.

The statutory approach should not be treated as perfect either. Laws can be clumsy, and they can be avoided or struggle to keep pace. Their strength is that they give people rights they can use.

The best systems tend to combine both: binding rules for the basic protections and voluntary standards for the fast-changing details. That is a design India could aim for.

The choice ahead

Artificial intelligence is arriving in offices, factories and public services at speed. The central question is not whether it will change work, but who will have a say in how. If the answer is only the companies that build and use it, then workers will be left to adapt as best they can. If the answer includes the law, then workers have a stake.

The week's two events show the options clearly. One path asks the powerful to restrain themselves. The other sets the limits in law. For a country with as many workers in the path of this change as India, the decision on which path to follow is not a technical matter. It is about fairness, and it will not wait for long.