Open letter: Brazil cannot rely on luck in the age of AI

To the candidates in this election, and to those already governing and legislating,

Artificial intelligence is already changing work, digital security and relations between countries. Brazilian hospitals, banks and public agencies depend on systems the country does not control. Cyberattacks may become harder to contain, while rules for the most powerful AI systems may be set without Brazil at the table.

The four commitments below are a proposal for candidates and elected representatives. A public response explaining which measures they support and how they intend to put them into practice would help voters assess their plans.

1. Defend verifiable limits for high-risk systems

Brazil should defend common rules for developing the most advanced systems, with independent verification that each country is complying. The rules should cover systems capable of causing severe harm, such as carrying out cyberattacks without human oversight, helping produce biological weapons or escaping the control of their operators. They should require safety evaluations, reporting of serious incidents and public criteria for suspending and resuming activities, without preventing the use of systems that pass those evaluations.

2. Work toward an international agreement open to other countries

Brazilian diplomacy should keep channels between the United States and China open and bring the issue to the UN, the G20 and BRICS. An agreement must allow other countries to participate, include Brazil in technical verification and guarantee public rules, a right to appeal and predictable access to permitted uses.

3. Prepare Brazil’s services and workers

Brazil needs to map which models, cloud services and suppliers its health care, energy, banks, communications and public services depend on. Each sector should test how it would keep operating if a supplier failed or cut off access. Exceptional powers to counter cyberattacks should have time limits, judicial oversight and public reporting.

The government should also use official data to track the employment and income of people entering the labor market. Training programs should demonstrate that participants gain employment compared with people who did not take part. Temporary income protection must have defined funding.

4. Turn energy into capacity available in Brazil

Brazil should turn its energy advantage into data centers, grid connections and computing services operated in the country. Companies, researchers and public agencies need to be able to contract this capacity to run models and maintain essential services, including if access to foreign suppliers is restricted. REDATA and other incentives should serve this goal, with public costs and obligations that can be verified.

Grid upgrades should be paid for by those who benefit, not through electricity bills. Brazil should also offer stable rules, fast imports of equipment it does not produce and room for domestic suppliers where they are competitive. Anyone who receives an incentive without meeting its obligations should return the benefit. Incentivized capacity remains subject to the same risk limits Brazil defends abroad.

First steps

  • Candidates for the Chamber of Deputies, the Senate or a State Legislative Assembly: if elected, request or support a public hearing by June 30, 2027, on AI risks, service continuity and access to computing.
  • Those already serving in Congress: use a formal request for information to ask the executive branch for a plan, with responsible parties, deadlines and costs, to protect essential services and expand AI capacity available in Brazil.
  • Candidates for the presidency or a state governorship: in their first hundred days, designate a responsible official and begin an inventory of essential services that depend on cloud and AI suppliers.

By signing this letter, each signatory affirms that Brazil can matter in the age of artificial intelligence: protecting its services, turning energy into computing capacity and taking part in international decisions about AI. That path depends on choices made now.

Brasil 2030 · independent research, without funding · who we are

Sign the letter

To prevent fake signatures, we ask you to sign in with your Google account. Google tells us your name and email address. The email is never published. How we handle your data.

Signing means calling for these commitments. It is not an endorsement of any candidacy, party or supplier, and an individual signature does not represent any institution.

Signatories

4 signatures

  1. Luiz PicciniCo-author · Head of Product at Tela
  2. Danilo NaiffCo-author · machine learning researcher
  3. Pedro CastilhoCo-author · CTO at Cumbuca
  4. Ivan M. FrancoCo-author · communicator and organizer at Effective Altruism Brazil