Superintelligence and Copyright: What Your Company Should Check

Superintelligence and Copyright: What Your Company Should Check

Check authorship, tool terms, licenses and personal data before publishing AI-generated content. A checklist that treats copyright as still open.

•

Want to apply this in your company?

Book a call with Juan →

When a company publishes text, images or audio generated by AI, the copyright question does not end with the prompt sent to the model. The team still needs to check who can be recognized as the author, what the tool's terms say about the output, whether the inputs used were licensed, and whether the content contains personal data. The question of authorship for machine-generated work has no settled answer in Brazil or in most other countries. The word superintelligence does not change this calculation, because the practical risks come from the models your company already uses today.

Key takeaways

  • Authorship of content produced without a meaningful human contribution is a gray area. Treat it as a risk to manage, not as a certainty.
  • The tool's contract matters as much as the law. Read the clauses on ownership of outputs, use of your inputs, and responsibility when third parties raise complaints.
  • Photos, text excerpts, sources, music and trademarks have owners, and AI does not transfer those rights to the person who uses the result.
  • When content involves customers, staff or testimonials, personal data rules must be reviewed alongside copyright.
  • A recorded human approval and an inventory of what was generated with AI are the foundation for answering any question later.

Why authorship is still open

The copyright tradition, including Brazilian law, is built around works created by people. When a model produces a text from a prompt, the question becomes how much of the creation was human: the choice of subject, the structure requested, the final editing, the selection among versions. These factors tend to matter, but the degree of participation needed for protection is still debated, and the answer can vary with the type of work and the specific facts.

In our work with companies, the practical point is not to win that debate. It is to be able to show who did what, and when. This is a working hypothesis, not a legal conclusion. For that reason, a company should treat ownership as something to confirm case by case, not as a presumption it can rely on without checking.

What to check in the tool's terms

Before a tool is made available to a team, read the contract or terms of use for the version actually being paid for or used, not only the marketing page. The points that tend to be most useful in this review are:

  • Which plan is in use and whether it covers the whole team, since personal accounts often have different conditions from business accounts.
  • Who holds ownership or usage rights over the generated output, and whether the clause grants those rights broadly or narrowly.
  • Whether inputs, including prompts, files and conversations, can be used to train models, and whether there is a setting to turn that off.
  • How long data is stored and where it is processed.
  • What the provider takes on, if anything, when a third party complains that the output infringes its rights.
  • Whether the company has an internal policy that defines what can and cannot be done with these tools, as the company policy for employee ChatGPT use describes.

Terms change. A review done once and forgotten is only useful until the next update, so the copy you read should be saved with the date you consulted it.

Inputs, references and third-party licenses

AI does not clean up the origin of material. Asking for a result "in the style of" a specific artist, pasting a protected excerpt into a prompt, using a stock photo without a commercial license, or reproducing a competitor's visual identity all remain questions for the person who uses the output. A simple input check can include:

  • Verifying the license for each image, font, track or audio file, including whether it allows commercial use and not only viewing.
  • Confirming that excerpts pasted into prompts from external sources may be reused.
  • Preventing third-party brands, logos and characters from appearing in material without authorization.
  • Avoiding requests that copy the identity of living artists or authors.
  • Requiring written authorization before publishing materials that belong to clients, including their logos, photos and product designs.

The same discipline applies to reference images. A mood board that a designer uses to guide a human-made piece is different from a reference the model is told to imitate closely, and the team should be able to explain which situation it is in.

Personal data in the same content

A marketing piece or an automated reply can carry personal data without anyone noticing: a customer's name inside a testimonial, an employee's face in a team photo, a service history pasted into a prompt. In these cases, the review needs a clear legal basis and a purpose compatible with what was originally collected, under the LGPD. If the data passes through providers outside the country, that adds another layer of attention, and the article on international data transfer under LGPD when using AI providers covers that point in more depth. The Brazilian data protection authority, ANPD, publishes general guidance on its official website, which is a sensible starting point for the team responsible for the review.

A practical habit helps here: before sending a prompt that includes a real customer record, ask whether the task could be done with the data removed or replaced with placeholders. Often it can, and the content loses nothing while the exposure drops.

A five-step review plan

A company that does not yet have a routine for this can start with a short sequence. The goal is not perfection on the first pass, but a repeatable process that someone is accountable for.

  • Inventory which published content or content sent to customers involved AI, separated by channel, such as website, social media, email, WhatsApp or printed material.
  • Classify each content type by risk. Institutional text, product imagery, testimonials, legal notices and material containing personal data each need different levels of review.
  • Read the terms of the tools in use and save the version you read, with the date you consulted it.
  • Name a human owner who approves each content type before publication. The approver should have enough knowledge to spot a problem, not just a signature to add.
  • Record the origin of each piece: the prompt or method used, the tool, the reviewer, the sources of the inputs and the approval. This record is what allows you to answer a complaint with facts instead of memory.
  • Repeat the review when a tool changes its plan, its terms or the way it is used inside the company.

The first two steps usually reveal more than expected. Many teams discover that AI is already in use in places nobody formally approved, which is an operational problem before it is a legal one.

Illustrative scenario

Illustrative scenario: a fictional home decor shop uses an image generator to create campaign banners. The team asks for a style that resembles a well-known illustrator and includes the logo of a partner brand in one of the designs. During review, the company replaces the style request with its own art direction, removes the partner logo because permission was never given, and keeps a record of who approved each banner. If a complaint arrives, there is a history showing what was made, by whom and with which inputs. The shop does not need to win an argument about authorship to respond responsibly, but it does need to be able to show its process.

How we apply the method in practice

Our method starts by mapping where generated content enters the business process. Then we limit which tools and which types of material are allowed. We run a real workflow with logging and human approval, and only after that do we hand the routine to an internal owner, who decides whether to expand, adjust or stop the practice. This approach does not replace legal advice. When there is doubt about ownership or liability, the case should go to a lawyer who knows the specific tools, contracts and sector involved. For a view of how we work with companies, see the page on AI consulting in Brazil.

Frequently asked questions

Does a company own the content its AI generates? It depends on the tool's contract and on the law that applies, and authorship of a work without a meaningful human contribution is not settled. Read the terms of the version you use and treat ownership as a point to confirm, not as something to assume.

Can I use an AI-generated image in an advertisement? The contract may allow commercial use, but the image should not reproduce a third party's brand, person or protected work. Review the output before publishing and keep a record of the approval.

Is it enough to add a notice saying the content was made with AI? A notice supports transparency, but it does not replace licenses for inputs, image authorization or a legal basis for personal data. It does not correct material the company had no right to use.

Does this change when superintelligence arrives? There is no consensus on what superintelligence would be or when, if ever, it would exist. The questions about contracts, inputs and data apply to the models available today, and the best preparation is to keep processes, owners and review records clearly defined now.

Sources

copyrightgenerative AILGPDAI governancecontent review