AlphaJuri

A αlphajuri compra o RPV de honorários que você já ganhou. Análise em minutos, proposta em até 48h e dinheiro na sua conta em até 72 horas. Sem dívida, sem cartório, sem regresso.

Article

AI in Legal Practice: the 86.6% Error Fear, in Context

The figure circulating in search results says most Brazilian lawyers fear errors in AI output. Here is what can actually be verified, where machine answers break in legal work, and why the review — and the payment date of an RPV — stays human.

alphajuriOctober 20, 2026 · 6 min read

In short

Artificial intelligence in law practice is the use of generative and analytical models to draft, summarise and search legal material, under a professional rule that keeps a human lawyer answerable for every output that leaves the office — which is why surveys on adoption keep measuring fear of error, not fear of the tool.

Stale data breaks answersThe MEI revenue ceiling has stood at R$81,000 a year since 2018 with no readjustment ^[7]; a model repeating an earlier marker is confidently wrong.
Status is not textWhen the STJ took Tema 1.454 under the repetitive-appeals rite, cases on the same question were stayed in the second instance ^[6] — no static model tracks that.
Assignment, not lendingUnder art. 295 of the Civil Code, the assignor answers for the existence of the receivable at the time of assignment, not for the debtor's solvency ^[11].
Timings are targets, not magicAnalysis in minutes, proposal within 48 hours, money in the account within 72 hours — after the file is checked.

The 86.6% figure, in context

We did not confirm the 86.6% figure in a primary AASP document, so it is not stated here as a verified number. The query behind this page is simple: a survey attributed to the Associação dos Advogados de São Paulo (AASP) has been reported as showing that around 86.6% of lawyers fear errors in artificial intelligence output, while adoption keeps growing and human review stays mandatory inside the office. That is the direction of the coverage, and it matches what anyone who has asked a model about a running case already knows.

The AlphaJuri editorial rule is the same one we apply to a case file: a number we have not read in the primary document does not get stated as if we had. So this article does what the search result around it does not — it shows, with sourced examples, exactly how an AI answer about Brazilian legal and fiscal matters goes wrong, and what that means for a lawyer waiting on fees already recognised in an RPV.

The percentage appears in press coverage of an AASP survey, not in a document reproduced here.
The concern reported is about reviewing output, not about refusing the tool.
A model answers from the data it was trained on, and that data ages.
No AI answer changes the official payment calendar of an RPV.

Three ways an AI answer breaks

The failure is rarely dramatic. It is a date, a procedural status, a bill treated as if it were law. Three verifiable examples, all from the last cycle of Brazilian legal and fiscal news:

Dates driftThe MEI revenue ceiling was raised to R$81,000 a year in 2018 and has had no readjustment since ^[7] — a model that places the marker in another year is wrong about something a reader can check in one line.
Status changes silentlyThe STJ affected Tema 1.454, on whether the dispensation of Fazenda Nacional fees under art. 19, § 1º, I, of Law 10.522/2002 is limited to items I to VII, and determined that second-instance cases on the same question be stayed ^[6]. Nothing in the statutory text announces that.
Approved is not in forceThe Chamber of Deputies approved raising the MEI ceiling to R$110,000 in 2027 and R$140,000 in 2028 by 461 votes to 4, and the proposal still depends on the Senate and on presidential sanction ^[8].

A model does not know what changed yesterday; the professional who signs does.

Human review is the deliverable

The survey figure is usually read as resistance. It is closer to professional hygiene. In a law practice, the output is never the draft — it is the signed document, and the signature carries responsibility that no vendor absorbs.

That is why the useful question is not "is AI reliable". It is "what in this answer is checkable, and against what". A summary of a procedural rule is checkable against the rule. A claim about the status of a repetitive-appeals theme is checkable against the court's own communication, as in the stay determined under Tema 1.454 ^[6]. A number about a revenue ceiling is checkable against the law in force ^[7] and against what is still a bill ^[8].

Applied to fees, the discipline is identical. A model can summarise what an RPV is. It cannot tell you whether this specific receivable, in this specific case, is eligible for assignment today. For that, somebody has to open the file.

AI does not move the payment date

Here is where the conversation usually stops being abstract. The lawyer has fees recognised in a Requisição de Pequeno Valor. The decision is final, the amount is defined, and the payment follows the public entity's own calendar. No model, no prompt and no automation shortens that queue.

What the lawyer can decide is whether to wait for it. An RPV of recognised fees is an asset that already exists in the file, and it can be assigned. That is a different operation from waiting — and a different operation from borrowing, which is not what happens here.

If the distinction between the two instruments is not settled in your head, the difference in amount, rite and waiting time between precatório and RPV is the first thing to read, because it determines how long the official wait actually is.

What the file has to prove

The assignment of a receivable is governed by arts. 286 to 298 of the Civil Code. Under art. 295, in an onerous assignment the assignor answers to the assignee for the existence of the receivable at the time of the assignment; responsibility for the debtor's solvency is not automatic and depends on an express clause, under art. 296 ^[11]. In plain terms: what the law secures is that the asset exists, not that the public entity pays faster.

That is the legal spine of the operation. AlphaJuri analyses the eligibility of the receivable before any proposal — the file is checked first, and only then priced. It is also why the paperwork is modest: send the case number and the OAB registration through the contact page, and from there AlphaJuri verifies eligibility and presents a proposal. By AlphaJuri's own service information ^[9], the analysis is free, carries no commitment, and a specialist responds within 48 hours.

No notary. No recourse. No ongoing obligation left on the office's books. The receivable changes hands; the lawyer does not acquire one.

What AlphaJuri does

AlphaJuri buys the RPV of fees the lawyer has already won. It is a purchase of the receivable — not a loan, not a notary procedure, not recourse against the office. The types of judicial receivable covered include precatórios, RPVs and sucumbência fees recognised in court ^[10]. AlphaJuri analyses eligibility before any number is put on the table.

The timeline is the argument, and it is the target the operation works to, case by case: analysis in minutes, a proposal within 48 hours, money in the account within 72 hours — always after the file itself has been checked. You can see the sequence step by step on how it works.

Artificial intelligence will keep changing how a petition is drafted, and the survey circulating this month suggests the profession has already decided to keep a human reading the output. The same standard applies here: a model can summarise what an RPV is, and a specialist still opens the file before anything is priced. What is signed at the end is an assignment — a purchase of the receivable, not a loan — by people who answer for it.

FAQ

Does the 86.6% figure come from an official AASP document?

We did not confirm it in a primary document. The percentage appears in press coverage of a survey attributed to the Associação dos Advogados de São Paulo, so we report the direction of the finding — high adoption of AI alongside a strong concern with errors and with keeping human review — without restating the decimal as a verified fact of our own.

Why do AI models get Brazilian legal and fiscal facts wrong?

Because they answer from data that has a cut-off date. The MEI ceiling has been R$81,000 a year since 2018 with no readjustment ^[7], while the increase to R$110,000 in 2027 and R$140,000 in 2028 was approved by the Chamber and still depends on the Senate and sanction ^[8]. Procedural status is worse: the stay of second-instance cases determined under STJ Tema 1.454 ^[6] exists in a court communication, not in the statute a model has memorised.

Can artificial intelligence speed up the payment of an RPV?

No. The payment follows the public entity's calendar, and no software changes it. What exists is the option of assigning the receivable: AlphaJuri buys the RPV of fees already recognised, with analysis in minutes, a proposal within 48 hours and money in the account within 72 hours once the file is checked.

Is assigning fees the same as taking on an obligation?

No. It is an assignment of a receivable under arts. 286 to 298 of the Civil Code. Art. 295 establishes that, in an onerous assignment, the assignor answers for the existence of the receivable at the time of the assignment, while responsibility for the debtor's solvency depends on an express clause under art. 296 ^[11]. There is no notary step and no recourse against the office.

What does AlphaJuri need to analyse an RPV?

The case number and the OAB registration. With those, eligibility is verified and a proposal is presented. By AlphaJuri's own service information ^[9], the analysis is free, carries no commitment, and a specialist responds within 48 hours.