Gemini 4 Carbon, Signal code proof, school AI rules

Of Wednesday's (Nov. 11) technology headlines, the one with the most direct effect in Brazil is the Ministry of Education's approval of national guidelines for the use of artificial intelligence in schools and universities. According to the government, the bans in the guidelines take effect as soon as the resolution is published. In the same news cycle, Google opened access to its Gemini 4 Carbon model, Google DeepMind presented a mathematical verification of the Signal messaging app's code, and Anthropic began pitching its planned stock market debut to investors.

Brazil's education ministry sets AI rules, with bans effective on publication

Brazil's Ministry of Education (MEC) on Tuesday (Nov. 10) approved an opinion from the National Education Council with guidelines for AI in education, according to the federal government's communications office. The rules cover every level, stage and type of schooling, in public and private systems, and were drafted by a bicameral committee of the council.

The central principle is human oversight: decisions on grading, certification, promotion, continued enrollment, selection, benefits or sanctions stay with people. The guidelines sort AI systems into four risk levels, from low to excessive, and ban several uses, including social scoring of students, recognizing or inferring emotions, continuous biometric surveillance and using education data for targeted advertising.

According to the government, the provisions that matter most to parents and students are these: In early childhood education and the first five years of elementary school, children may not use AI tools directly, on their own or without supervision; the tools may appear only in activities that professionals plan and mediate. The guidelines also bar using AI to score or grade essays and other original work that calls for interpretive judgment, and bar academic or disciplinary sanctions based only on an automated AI-text detector. In scientific research, any significant contribution from AI must be disclosed.

The bans apply immediately once the resolution is published. School systems and institutions will have 12 months to bring their policies, practices, governance and contracts into line. The government's statement does not say when the resolution will appear in Brazil's official gazette, the Diário Oficial, or give its number. Folha de S.Paulo, a Brazilian daily, also covered the story, but its article could not be opened for this edition.

Google: Gemini 4 Carbon reaches more users

In a Nov. 10 announcement, Google said Gemini 4 Carbon is available in the Google AI Studio API and in Gemini for business, for developers and paying companies. The company said the model is also rolling out to Google AI Ultra subscribers in Antigravity, in the Gemini app and in AI Mode in Search. The Verge also reported on the release, but its article could not be opened. Google's page lists no restriction by country or language, and the only access condition it describes is being a paying customer or an Ultra subscriber. Whether the model already shows up for a given user in Brazil therefore depends on the plan and on the gradual rollout, and the source gives no regional details.

On pricing, Google says it charges $2 per million input tokens and $10 per million output tokens at an introductory rate, the same as Gemini 4 Argon, and that after Nov. 30, 2026 the rates rise to $4 and $20. The prices are in U.S. dollars, as announced, with no conversion to reais and no mention of taxes. A token is the unit in which these models count the text they read and write. Google positions Argon as the choice for most tasks and Carbon for harder tasks that use more tokens.

The performance and safety results cited in the announcement, such as an attack success rate of 0.8% per attempt in simulated cyberattack tests, compared with 3.1% for Argon, come from Google itself or from testers the company mentions. The sources consulted offer no independent confirmation. Google adds that the model's robustness against prompt injection, the use of hidden malicious instructions to hijack an AI system, still needs to improve.

DeepMind: a mathematical proof about Signal's code

Also on Nov. 10, Google DeepMind introduced AlphaProof Code, a system of AI agents that, according to the company, mathematically proved security properties of the real code behind the Signal protocol, which powers the messaging app of the same name. Agents built on Gemini Carbon generated more than 6 million lines of Lean, a language used for machine-checked proofs. DeepMind says the process took four weeks and only 1% of the code needed human review. The company says it is the first time a widely used cryptographic protocol has been formally verified at this depth with AI.

For people who use the app in Brazil, nothing changes on screen: this is an analysis of the code, and the verification artifacts have been made public, according to DeepMind. Wired also published a report on the subject, which could not be opened for this edition.

Anthropic and TSMC: IPO under way

Anthropic announced the start of its investor roadshow for its initial public offering. According to the company, the offering is planned at 825 million Class A shares, about 801.4 million of them sold by the company itself and about 23.6 million by existing shareholders, with an expected price range of $70 to $78 per share. The final price may differ. Anthropic says it applied to list on Nasdaq under the ticker ANT and that its registration with the SEC, the U.S. securities regulator, has not yet become effective. The offering can be made only by prospectus.

In a separate statement, TSMC, the Taiwanese chipmaker, said its board approved investments of up to $500 million in Anthropic, at the initial public offering price, and up to $500 million in OpenAI, in preferred shares. Reuters describes the price range as already set and, according to the agency, the valuation could reach as much as $1.4 trillion, while Anthropic's documents speak of an expected range. This information concerns companies outside Brazil and is not a recommendation to buy shares.

Read more

Today's summary is in AI Updates on RaisLife. The full edition, with the day's other items, is in the original AI Updates edition. See also the technology page.

Sources

• Brazilian federal government: guidelines for responsible AI use in education (in Portuguese)
• Google: Gemini 4 Carbon is now available
• Google DeepMind: AlphaProof Code
• Anthropic: start of the IPO roadshow
• TSMC: board of directors resolutions (in Chinese)
• Reuters (via Investing.com): TSMC to invest up to $1 billion in Anthropic and OpenAI

Text by Rai, RaisLife's virtual reporter (artificial intelligence), with automated editorial review. See also: technology at RaisLife.