AI Chiefs Promise to "Self-Police" in Washington, While California Writes Rules for Workers

AI executives agreed at the White House this week to self-police their industry, while California's governor signed laws on AI job losses and workplace surveillance. The two approaches show a widening gap between voluntary promises and binding rules. India, which has neither, should pay attention.

Two responses in the same week

On Tuesday, President Trump said top technology leaders had signed a voluntary accord to "self-police" AI development. The executives reportedly agreed after several high-profile mishaps and warnings from some of their own colleagues about the technology's dangers.

On Wednesday, California Governor Gavin Newsom signed laws aimed at protecting workers from AI-related threats, including potential job losses and workplace surveillance. Earlier this month, he signed a law requiring operators of AI chatbots to carry out risk assessments before rollout. He also signed an executive order requiring the state to consult experts to improve oversight of the industry.

One approach asks companies to promise. The other tells them what they must do.

A fight over words

The politics go beyond policy. Newsom also signed an executive order requiring state agencies to keep calling the technology "artificial intelligence". That was a response to President Trump's recent order directing US diplomats to use the term "super intelligence". The labels matter because they frame the policy question. "Artificial intelligence" suggests a tool to be managed. "Super intelligence" suggests an entity beyond human control, and invites either alarm or inevitability.

Newsom sharply criticised Washington for not passing comprehensive federal AI regulation while industry leaders warn about risks. "We have to do a lot more in the absence of federal leadership," he said, and he did not rule out a special legislative session. It was his last day to sign or veto bills before he leaves office in January, so the timing carries political weight, with midterms weeks away.

What voluntary pledges can and cannot do

Voluntary accords have a place. They can set norms quickly, and they cost little to adopt. Industry knowledge also matters, since regulators often lag technical reality. But their limits are well known:

• They are not enforceable. There is no penalty for failing to follow them.

• They are written by those they bind.

• They rarely cover workers, who are the people most affected by deployment decisions.

• They can reduce pressure for binding law by creating the appearance of action.

The reported trigger, several high-profile mishaps, suggests that voluntary measures follow harm rather than prevent it.

What California's approach does

California's laws target two concrete risks.

• Job loss: Rules on AI-related displacement give workers a stake before deployment, not after.

• Surveillance: AI-driven monitoring of employees, from productivity scoring to behavioural tracking, can intensify work and erode privacy without workers knowing the criteria.

Pre-deployment risk assessments for chatbots add a third layer, requiring companies to identify harms before launch.

The fair criticisms are also worth stating. A patchwork of state laws can raise compliance costs and create inconsistent standards. Companies argue that heavy rules may push investment elsewhere. Some policy analysts say job-loss rules are hard to enforce because causation is difficult to prove. These are real trade-offs, not excuses, and the evidence on how California's laws perform will take time.

The money behind the debate

The industry's financial stakes explain the intensity. OpenAI is reportedly looking to raise $30 billion at a $1.4 trillion valuation, and its CEO has said it will not go public until its models are safe. A separate frontier lab's IPO prospectus, reported this week, shows deep dependence on Big Tech partners. Broadcom is reportedly assembling $60 billion to fund chips for one lab.

When valuations reach these levels, voluntary commitments sit alongside enormous commercial pressure to deploy fast. That is exactly the situation in which binding rules matter more.

Why India should care

India is both an AI user and a major AI workforce. Three points stand out.

First, workers. India's IT services, BPO and back-office sectors employ millions in tasks that automation can reach. The question of who protects workers from AI-driven displacement and surveillance is not hypothetical.

Second, law. As far as we know, India relies on general data-protection and IT law rather than a dedicated statute on AI in the workplace. The India AI Impact Summit in February focused on citizen-centric solutions for the Global South, but summit declarations are not enforceable rules.

Third, capital. The Finance Ministry has said India's limited participation in global AI developments weighs on its ability to attract investment. India is rolling out AI in governance too, such as the ₹1,789.52 crore AI-enabled system to manage Delhi's traffic that the Cabinet cleared. Public deployment raises its own accountability questions about transparency, error handling and data use.

India does not need to copy California. It does need a position on who is responsible when AI changes someone's job.

What a workable Indian approach would include

• Workplace disclosure: Employers should tell workers when automated systems monitor or evaluate them, and on what criteria.

• Impact assessments for large deployments: Particularly in public services and large employers.

• Transition support: Reskilling funds tied to measurable displacement, not general announcements.

• Redress: A clear channel for workers to challenge automated decisions.

• Data on impact: Regular publication of AI-related job changes by sector.

Government's strengths here include a track record on digital public infrastructure, which shows it can build systems at scale. The test is whether it can regulate them with the same ambition.

Bottom line

The week showed two models. Washington is relying on industry promises, and California is writing enforceable rules. Neither is perfect, but only one gives workers recourse. As AI spreads through Indian offices, call centres and government systems, India will have to decide which model it follows, and soon.