AI Regulation in 2026: The EU, Colombia and Brazil in September

This article is for informational purposes only and is not legal advice. Consult a lawyer about your specific situation.

If your company operates in Colombia or Brazil and also sells software or services to Europe, AI regulation is no longer a “later” topic. But it isn’t a single block either: each jurisdiction moves at its own pace, and in September 2026 the map looks very different from what was expected at the start of the year.

Europe adjusted its deadlines but kept obligations that already apply. Colombia has a new bill on the table. Brazil paused the vote on its legal framework until after the elections.

This article sums up where each one stands and what you can do today so you don’t have to improvise tomorrow.

European Union: the Digital Omnibus on AI is in force

The European Union amended its AI Act through the so-called “Digital Omnibus on AI.” The Council gave its final approval on June 29, 2026; it was published in the Official Journal on July 24 as Regulation (EU) 2026/1744 and entered into force on July 27.

The main changes:

Stand-alone high-risk systems (Annex III): application is delayed from August 2, 2026 to December 2, 2027.
AI in regulated products (Annex I): delayed to August 2, 2028.
Article 50 transparency obligations: August 2, 2026 stays. That makes it a live compliance date. There is a grace period until December 2, 2026 for watermarking of systems already on the market.
New prohibition: AI systems that generate non-consensual intimate imagery or child sexual abuse material, with compliance required by December 2, 2026.
AI literacy: the obligation is softened, moving from ensuring staff AI literacy to supporting it.
Regulatory sandboxes: the deadline moves to August 2, 2027.

What it means for a Latin American company

The high-risk delay buys time, but it isn’t a general pause. If your AI product is used or sold in the EU, it’s wise to assume transparency is already enforceable and that the December 2026 clock is ticking for watermarking and the new prohibition. Check with legal counsel whether and how these obligations apply to you, depending on your role and market.

Colombia: a new bill, not yet debated

On July 21, 2026, Pacto Histórico representatives filed Bill 025 of 2026 (Cámara) on artificial intelligence. According to its own statement of reasons, the previous bill (PL 043 of 2025 Senate – 324 of 2025 Cámara) was archived after failing to pass its first debate.

The most relevant points of the filed text:

Risk-based classification (Art. 5): includes a “Critical Risk” (“Riesgo Crítico”) category for systems not permitted for general use.
National authority (Art. 7): designates MinCiencias, Colombia’s science ministry, as the national AI authority.
Sanctions (Art. 32): fines of up to 3,000 SMMLV (monthly legal minimum wages), suspension of up to 24 months, and national blocking of a system.
Transition (Art. 34): systems already in operation would have to adapt.

The bill has not been debated yet and its outcome is uncertain. Even so, Article 34 sends a clear signal: if a law like this passes, complying with new systems won’t be enough; those already in production would need adjusting too.

Brazil: the legal framework waits for the elections

In Brazil, PL 2338/2023, which sets out the country’s AI legal framework, was scheduled for a vote in the Chamber of Deputies in May 2026, but the vote didn’t happen.

As reported by Mobile Time on August 24, 2026, rapporteur Aguinaldo Ribeiro said at Febraban Tech 2026 that work should resume after the October 2026 elections. Copyright remains unresolved. After the Chamber, the bill must return to the Senate, which approved it on December 10, 2024.

The map at a glance

Jurisdiction	Instrument	Status as of September 2026	Key dates
European Union	Regulation (EU) 2026/1744 (Digital Omnibus on AI)	In force since Jul 27, 2026	Art. 50: Aug 2, 2026 · Watermarking and new prohibition: Dec 2, 2026 · Sandboxes: Aug 2, 2027 · Annex III: Dec 2, 2027 · Annex I: Aug 2, 2028
Colombia	Bill 025/2026 (Cámara)	Filed Jul 21, 2026; not yet debated	To be determined
Brazil	PL 2338/2023	Pending in the Chamber; then returns to the Senate	Work expected to resume after the October 2026 elections

Readiness checklist for your company

None of these actions depends on any specific law passing. All of them leave you better prepared for any of the three scenarios:

AI system inventory. Which systems use AI, for what, with which data, and in which markets they operate.
Preliminary risk classification. Use the European categories as a reference; both the EU and the Colombian bill follow a risk-based approach.
Transparency today. Check whether your products used in the EU already meet Article 50 obligations.
Watermarking before December 2, 2026. If you generate synthetic content for European users, plan the marking now.
Prohibited content controls. Confirm that no system can generate non-consensual intimate imagery or child abuse material, with evidence of the safeguards.
AI literacy program. Even though the obligation was softened, training your team is still the foundation of responsible use.
High-risk roadmap. If any system could qualify as high-risk, use the runway to December 2027 to document, test and supervise it.
Adaptable design. Build with logs, traceability and room to adjust, keeping in mind that a law may require existing systems to adapt.
Owner and monitoring. Assign a person or committee to track Bill 025/2026 in Colombia and PL 2338/2023 in Brazil.

FAQ

Does the European delay mean I can wait until 2027? Not entirely. High-risk deadlines moved, but Article 50 transparency obligations keep the August 2, 2026 date, and both the new prohibition and the watermarking grace period end on December 2, 2026.

Does Colombia already have an AI law? No. Bill 025/2026 (Cámara) was filed on July 21, 2026 and has not been debated yet. Whether it passes, and its final text, are uncertain.

What should I do about Brazil in the meantime? Follow the process and advance the practices that work under any framework: inventory, risk classification, traceability and human oversight.

Conclusion

AI regulation in 2026 isn’t moving in a straight line: Europe is adjusting deadlines without dropping obligations, Colombia is trying again with a new bill, and Brazil is waiting for the electoral cycle to pass.

For a Latin American company, the best strategy isn’t to wait for everything to be settled, but to build traceable, documented and adaptable systems starting now. If you want to design AI solutions ready for that landscape, explore our AI for business service or get in touch.