THE SIGNAL IN ONE SENTENCE

Brazil has signed the Antalya Artificial Intelligence Pledge at COP31 in Türkiye. That sentence now has more substance than it did when the pledge was still a proposal. There is a final text. There are four pillars. There are voluntary commitment menus for governments and for companies, research institutions, cities and other non-government signatories. There is a plan to review collective progress at each UN climate conference through 2030, beginning with COP32. There is also a very large escape hatch. The pledge says its objectives are voluntary, non-prescriptive and create no new obligations. So the useful question is not whether Brazil signed. It did. The useful question is what the signature changes on the meter and the payroll. Brazil's Ministry of Communications announced the country's adherence on November 16. The ministry described a commitment to balance innovation, environmental protection and the future of work. It highlighted the electricity and water used by data centers, along with the effects of automation on employment. The ministry's telecommunications secretary, Hermano Tercius, said Brazil had emphasized protecting workers from job transformation and replacement caused by AI. That is a meaningful political statement. It is not the same thing as a worker-protection clause in the pledge. The final Antalya text focuses on four areas: sustainable power and infrastructure, responsible management across the AI lifecycle, assessment and reporting of environmental impacts, and support for AI applications that address climate problems, especially in developing countries. Its government annex includes an optional commitment to train or upskill people in sustainable AI and climate applications. It does not require signatories to preserve jobs, disclose layoffs, fund income support, bargain with workers or prove that training led to a comparable job. Brazil has added the employment promise in its own announcement. Brazil should now add the evidence. Build one public implementation ledger with two linked sides. The first side follows the meter. For every large AI or data-center project supported, licensed, procured or incentivized by the federal government, publish the facility, operator, location, project stage, electricity demand, hourly load profile, water source, cooling method, backup generation, grid connection, transmission upgrades, public subsidy and expected operating date. Report absolute consumption as well as efficiency. A new server can use less electricity per calculation while the facility's total load rises dramatically. A cooling system can improve water efficiency per unit of computing while a larger campus draws more water overall. Efficiency without scale is a flattering fraction with the denominator cropped out. Show the time and place of the electricity too. An annual renewable contract does not prove that a data center used low-emission power during the hour it stressed the local grid. Publish hourly consumption and the generation mix serving that load. Identify whether clean generation is new, whether storage covers the difficult hours and what happens during drought, heat or transmission constraints. Then show who pays. The pledge specifically says infrastructure costs associated with data-center development, including transmission upgrades, should be allocated appropriately. That is unusually useful language because the wires are where a grand announcement often becomes a household bill. Brazil should publish the substation, transmission, generation and water-system upgrades required by each major project, the party responsible for each cost and any expense shifted to electricity or water customers. If public money lowers the operator's cost, show the public benefit purchased in return. That may include a firm clean-power requirement, local training, reusable waste heat, shared grid equipment, water limits, community investment, transparent reporting or a right to reduce load during an emergency. The benefit should be an enforceable term, not a ribbon-cutting adjective. The second side follows the payroll. Do not reduce the employment effect of AI to a national count of jobs created and jobs lost. That arithmetic can call a transition successful while one occupation, city or age group absorbs the damage and another receives the opportunity. Report changes by occupation, industry, place, contract type, wage band, race, gender, age and disability where privacy and statistical quality allow. Separate four different events: An entire job disappears. A job remains but important tasks are automated. A job grows because AI creates new work. A worker is moved, retrained or hired into a different role. Those are not interchangeable outcomes. If a call-center worker keeps the same title while the system handles routine conversations and the human receives only angry, unusual or traumatic calls, the job has changed even if the payroll count has not. If a graphic worker uses a model to produce more assets under a tighter quota for the same wage, productivity has changed but the benefit did not necessarily reach the worker. If a public agency buys an automated decision tool and freezes entry-level hiring, the job effect can arrive without a layoff notice. The ledger should capture those cases. Start before displacement. Employers receiving AI-related public funding, tax support or procurement revenue should publish an employment impact assessment before a major deployment. It should name the affected tasks and roles, the expected timeline, the people consulted, the training budget, paid learning hours, placement commitments, wage protection, appeal path and the result that would pause the rollout. Workers and unions need access to the evidence early enough to change the plan. A consultation held after the software contract is signed and the head-count target is announced is an exit interview wearing a workshop badge. Training must be measured by what happens after training. Count enrollment, completion, certification, placement, wage change and retention after six and twelve months. Report who could not participate because the course was unpaid, remote, inaccessible or scheduled around a shift. Identify whether the new job is permanent, outsourced or temporary. The Antalya pledge's skills commitment can support useful programs. It should not become a ceremonial total of people who watched a webinar. Connect the two ledgers. Data centers create construction, electrical, cooling, security, maintenance and operations work. They can also draw scarce engineers and electricians away from public infrastructure or smaller employers. Hardware reuse and repair can create local work while reducing waste. Clean-power and grid investments can expand skilled trades. Automation inside the facility may limit permanent staffing after construction ends. Publish the difference between peak construction employment and steady operating employment. Show local hiring, apprenticeship completion, contractor wages, workplace safety, union coverage and the durability of the jobs after the launch photos are archived. The pledge's environmental annex gives Brazil a practical starting point. Governments can choose to request that AI providers disclose energy and environmental data. They can compile a public information and communications technology energy and greenhouse-gas inventory. They can move toward low-emission power requirements and adopt procurement rules for energy, water, cooling, durability, repair, reuse, recycling and electronic waste. Those are options, not proof that Brazil selected them. The ministry announcement does not identify which annex commitments Brazil checked. It does not publish a baseline, target, reporting date, budget, enforcement owner or national implementation plan. Publish the signed selection. Then turn each checked box into a row with a baseline, metric, target, deadline, responsible institution, data source, audit method, budget and current status. For data centers, useful rows include total and hourly electricity, carbon intensity, water withdrawals and consumption, water stress, refrigerants, backup fuel, embodied emissions, equipment life, repair, reuse, recycling, grid costs and emergency curtailment. For workers, add tasks changed, roles affected, consultation, paid training, placement, wage movement, working conditions, layoffs, hiring, contractor conversion and grievance outcomes. Name the missing data too. The pledge asks signatories to support standardized environmental measurement. Perfect standards will take time. Waiting for perfection can become an elegant way to publish nothing. Start with the information already collected by utilities, environmental agencies, employers, procurement systems, tax authorities and project developers. Mark estimates clearly. Publish the methodology and revision history. Improve the fields as international standards mature. Independent review matters. Operators should report the data, but they should not be the only people deciding what the data means. Give regulators, utilities, researchers, worker representatives, affected communities and public-interest groups access to the methods and enough underlying information to challenge the result. Protect personal information and genuine security details. Do not turn confidentiality into a blanket over public infrastructure costs, environmental demand or job effects. The annual COP review can create a rhythm. Before each conference through 2030, Brazil can publish a national Antalya implementation update. Show the commitments selected, projects covered, changes since the previous year, missed targets, corrective actions, public spending, grid and water impacts, employment results and unresolved disputes. The review should include one uncomfortable example. Name a project that changed because of the pledge. Perhaps the operator moved to a less water-intensive cooling design. Perhaps a utility refused an unfair cost transfer. Perhaps a procurement contract gained a repair requirement. Perhaps a rollout slowed until workers received paid training. Perhaps a facility did not proceed because the grid or watershed could not support it. If no project ever changes, the pledge is describing the future instead of governing the present. Brazil also highlighted its Computers for Inclusion program and the planned AI for Climate Change Institute during COP31. Those initiatives fit the pledge's themes of hardware reuse, digital access, local capability and climate applications. They still need their own evidence. Report how many devices remain in use, how repair extends useful life, what happens to unrecoverable equipment, which communities gain durable access, who controls the climate institute's research agenda, how local data is governed, which countries receive support and whether the tools improve water management, disaster preparation or environmental monitoring in practice. The pledge can be a useful international frame. Its final text recognizes electricity, water, land, materials, grids, infrastructure costs, measurement, developing-country access and annual review. That is considerably better than pretending AI floats in a clean cloud above physical consequences. But a voluntary promise becomes accountable only when the public can follow it through a meter, an invoice, a contract, a training record, a pay stub and a decision. Brazil signed the pledge. Now publish the operating ledger.

01

WHAT ACTUALLY CHANGED

Brazil announced on November 16, 2026 that it had formally joined the Antalya Artificial Intelligence Pledge during COP31 in Türkiye.

The COP31 Presidency and the International Telecommunication Union officially launched the final pledge on November 16.

The final text addresses sustainable AI infrastructure, lifecycle management, environmental assessment and reporting, and support for AI climate applications.

The pledge recognizes electricity, water, land, critical materials, electronic waste and grid-infrastructure costs as parts of AI environmental impact.

The pledge is voluntary and non-prescriptive and says it creates no new obligations.

A government annex offers optional commitments on provider disclosure, national energy and emissions inventories, low-emission power, data-center sustainability, hardware circularity, skills and climate applications.

A separate annex lets non-government signatories register numerical targets and dates for environmental disclosure, clean power, data-center requirements, hardware circularity, skills and climate support.

Signatories are invited to review collective progress at every UN climate conference through 2030, beginning with COP32.

Brazil emphasized protecting workers affected by AI-driven job transformation and replacement in its national announcement.

The final pledge includes an optional skills commitment but does not create a requirement to preserve jobs, disclose displacement or fund worker income protection.

02

WHY THIS MATTERS

Brazil has moved from discussing a draft pledge to signing a published voluntary framework with implementation options.

Voluntary language makes national targets, procurement rules, budgets and public reporting decisive.

Absolute energy and water demand can rise even when efficiency per calculation improves.

Hourly and location-specific data can reveal grid stress that annual renewable claims conceal.

Infrastructure upgrades can shift costs from data-center operators to households and other customers unless allocation is public.

Brazil's worker-protection statement goes beyond the pledge text and therefore needs a national implementation mechanism.

Job totals can conceal changes in tasks, wages, workload, location and employment quality.

Training counts are weak evidence unless programs lead to accessible, durable and fairly paid work.

A linked environmental and employment ledger can show the complete local consequence of an AI infrastructure project.

Annual COP reviews create a deadline for evidence but do not guarantee independent verification or consequences.

FIG. 673FROM SIGNATURE TO PUBLIC LEDGER
1PUBLISH THE COMMITMENTS→
2SET BASELINES AND TARGETS→
3METER POWER AND WATER→
4TRACK TASKS AND PAY→
5AUDIT COSTS AND OUTCOMES→
6CHANGE THE PROJECT
A voluntary pledge becomes useful when every commitment is attached to a measurable resource, a responsible institution, a worker outcome and a decision that can change.

03

WHERE IT COULD HELP

  • Publish which optional government commitments Brazil selected in the pledge annex.
  • Give every selected commitment a baseline, target, deadline, owner, budget, data source and audit method.
  • Report facility-level absolute and hourly electricity demand alongside efficiency metrics.
  • Publish water source, withdrawals, consumption, cooling method and local water-stress context.
  • Identify new generation, storage, substations, transmission and water infrastructure required by each major project.
  • Show which infrastructure costs are paid by operators, government, utilities and customers.
  • Make public subsidies conditional on measurable environmental and local benefits.
  • Require an employment impact assessment before publicly supported AI deployments.
  • Report job and task changes by occupation, place, contract type and wage band while protecting privacy.
  • Give workers and unions evidence early enough to change procurement and deployment plans.
  • Fund paid training before displacement and track completion, placement, wages and retention.
  • Separate peak construction jobs from permanent operations and maintenance work.
  • Include repair, reuse and responsible recycling in AI hardware procurement.
  • Publish methods, estimates, missing data and revision history while standards mature.
  • Give regulators, researchers, worker representatives and affected communities access to challenge the results.
  • Issue a national Antalya implementation update before each COP through 2030.
  • Name at least one project decision that changed because of the pledge.
  • Apply the same evidence requirements to Brazil's computing-reuse and AI climate-institute initiatives.

KEEP A HAND ON THE WHEEL

The Antalya Artificial Intelligence Pledge is voluntary, non-prescriptive and creates no new legal obligations. A signature does not itself establish a national rule, funded program, reporting duty or enforcement process. The government and non-government annexes contain selectable commitments. Brazil's announcement does not identify which optional commitments it selected or publish a completed annex. The pledge calls for progress reviews at future UN climate conferences, but its current text does not specify an independent verifier, penalty or remedy for missed commitments. Brazil's Ministry of Communications says the country defended worker protection and job maintenance. The final pledge text includes skills development but does not require job preservation, displacement reporting, bargaining, wage protection, income support or placement outcomes. The ministry announcement does not provide an energy or water baseline, target, implementation calendar, budget, responsible institution, worker-protection mechanism or national reporting format. Company and government environmental data may use different boundaries and methods. Comparisons require clear treatment of total demand, model development, inference, hardware manufacturing, networks, cooling, backup power and grid infrastructure. Watch for Brazil's selected annex commitments, a national implementation plan, facility-level electricity and water reporting, grid-cost allocation, public procurement rules, independent assurance, labor consultation, paid training and placement data, project-level changes, and the first COP32 progress review.

04

TERMS WORTH KEEPING

SOURCES AND VERIFICATION STATUS

This article was written from the materials below. Product claims and dates were checked against those sources on November 17, 2026.

THE PUBLICATION ENGINE

WANT A SIGNAL OF YOUR OWN?

We build source-grounded publications, private briefings, and editorial systems for organizations with something useful to say.

WORK WITH US