A

Aptus.AI

The metered unit is the AI conversation: Standard includes up to 50 AI messages a month and Professional sells use without limits. The assistant performs the research, document analysis, version comparison, drafting of filings and opinions, and the agents inside matter workspaces. Article 1.1 of the general terms describes the platform as making regulatory texts machine readable through artificial intelligence algorithms, and the pricing FAQ calls that patented conversion the key to letting the AI read legal documents.

Next-OS, launched 30 June 2026, adds iterative reasoning and actions in connected tools. Without generative models, Aptus.AI would still offer a curated database of more than 70 million provisions and 7 million decisions, automatic consolidated versions, daily alerts on the first working day after official publication, and Next Drive storage. That is the regulatory information service the company launched for financial compliance in 2021.

Aptus.AI built its own benchmark, LegalITA: 107 real tasks in civil, labor and tax law drawn from more than 17,000 Court of Cassation decisions, including 40 trap questions that presuppose a document never uploaded. A citation counts only when court, section, number and year are right and it supports the issue. The post of 2 September 2026 reports 87.0 percent of questions passed in full, 86.1 percent with at least one aligned citation and 93.3 percent recognition of missing documents; a July 2026 post compares eleven systems on the same test.

Three model judges score the results (Claude Sonnet 4.6, GPT-5.5 and Claude Opus 4.8), and one of them is also a compared system. Competitor products go unnamed, the task bundle is available on request, and the README says the citation grounding pipeline cannot be reproduced from the repository. Each source opens as a summary, full text and a link to the official publisher.

Article 7.1 of the general terms says the services are technical support that do not replace the team's users, who alone hold decision control. Article 7.2 places use under the user's own responsibility, and article 5.3 asks the user to evaluate every output carefully because content may be outdated or approximate. Next-OS reasons before producing an output and, in the vendor's words, does so on its own while showing the process in real time so each step can be followed.

When a request presupposes a document that was not uploaded, it flags this and asks for it before analyzing. The Studio Legale Spadoni case study describes a lawyer who verifies each output and keeps professional responsibility. The enterprise assistant page says the chat can send emails, create calendar events, tidy folders and run scheduled tasks, and no page says whether each action needs confirmation, what limits apply, or how a mistaken action is reversed.

The enterprise pages quote Enzo Battaglia of Banca Agricola Popolare di Sicilia, Donato Varani and Luca Gandolfi of RbyC, and Tommaso Tovaglieri of General Finance, and show the Intesa Sanpaolo logo. Coverage of the Next-OS launch on 8 September 2026 says Intesa Sanpaolo and Generali Investments use the platform. Case studies describe Avv. Daniele Spadoni of Studio Legale Spadoni in Latina, running contract, business crisis and civil litigation matters through workspaces, and Avv.

Fabio Francesco Franco of E-Legal in Rome and Milan, managing serial litigation for a public administration. The Rome Bar Association adopted the platform for its offices in October 2025. The published figures name no customer: 50 to 60 percent less time per opinion and over 1,300 hours a year for an unnamed large organization's legal team, and a DORA exit strategy check cut from three or four hours to about five minutes for an unnamed regulated fintech. No case study is dated, and none pairs a named customer with a measured result.

Article 5.4 of the general terms says the AI system respects data segregation criteria and carries out no specific training on data the user provides, without prejudice to Aptus.AI's ability to use that data to improve the services. The security page says customer data is never used to train AI models, and describes logical isolation between organizations and users and role based access in which collaborators see only what the account holder shares.

Article 14.3 binds both parties to keep confidential all information and documents exchanged and, at termination, to return or destroy them immediately with proof. Article 14.2 requires a data processing agreement where personal data is uploaded, offered as an addendum and not published. Secrecy under the rules of the legal profession, privilege and work product are never mentioned. The improvement use is not defined, and nothing states what the third party AI provider in article 3.3 may retain.

Article 1 of the general terms says the platform is not aimed at consumers as defined by article 4 of Legislative Decree 206/2005. The pricing FAQ says it is not designed for private users or people without legal or compliance expertise, and states plainly "Non forniamo consulenza legale." The free Academy course covers the rules on AI use in a firm, how the models work and where they fail, and a method running from the request to checking the sources, with a certificate the firm can keep as a documented measure.

The same plans are sold to accountants, labor consultants, public administration and compliance officers, who can draft opinions with no guidance on what a reader without legal training should do with legal output. The coverage is Italian, European and financial regulatory, while Spanish and German editions sell the same plans with no statement of where coverage stops.

Aptus.AI measures releases against its own LegalITA benchmark across civil, labor and tax law and publishes the evaluation code on GitHub, with the task bundle available on request. It reported scores for successive versions on 31 July and 2 September 2026, and says results stay stable across repeated runs and that the method's limits, including sample size and the need for stratified human validation, are declared in its white paper.

Article 5.5 of the general terms and the pricing FAQ state that the platform is neither a general purpose AI system nor high risk under the AI Act, on an assessment that is not published. According to the FAQ, an internal team of experts trains the algorithms. The governance documents that are published, the Model 231 organizational model, the code of ethics and the information security policies of 10 July 2024, say nothing about AI. No one is named as accountable for model behavior, and no result is broken out by practice area.

The privacy notice for the Aptus Next platform, last updated 1 July 2026, keeps data processed to provide the services for the contract term plus 12 months and security logs for the term plus six months. It says data is generally not transferred outside the European Economic Area. Article 14.3 of the general terms adds immediate return or destruction of confidential information at termination. The security page lists AES-256 encryption at rest and TLS 1.2 or later in transit, role based access on least privilege, logical tenant isolation, a penetration test at least once a year by external providers, and automated SCA and SAST code scanning.

The notice names cloud, AI and payment providers only by category, the data processing agreement is an unpublished addendum, and no breach notification period is stated; incident notification is something a DORA regulated client negotiates. The two exit rules differ: 12 months in the notice against immediate destruction in the terms.

Article 7.3 of the general terms says that, except for intent or gross negligence, Aptus.AI is not liable for any activity or decision taken through the platform or on its content, for any failure to achieve a result, or for any direct or indirect damage from using the platform. Article 9 says errors do not in themselves breach the contract. Article 11.2 has the client indemnify Aptus.AI for breaches of articles 4.2, 6.1, 8.1, 10.2 and 14.3.

Article 12 states that Aptus.AI holds professional civil liability insurance for the contract term with a limit of 5,000,000 euros, for Enterprise clients. The insured events, exclusions and any right of a client to claim are not stated, and the contract it backs leaves ordinary negligence and a wrong citation outside Aptus.AI's liability.

The Next-OS launch post of 30 June 2026 lists SharePoint, OneDrive and Outlook mail and calendar as live, with Google Drive, Gmail and Google Calendar announced. The enterprise assistant page says the chat can read and analyze documents and communications in connected tools and take actions such as sending emails, creating events and tidying folders and files. The Professional plan lists Google and Microsoft Office suite integrations, article 3.3 of the general terms names Microsoft Teams as an optional third party service, and documents can be revised in track changes inside the platform.

No documentation describes the permissions a connector takes, whether anything writes back to SharePoint, what an administrator configures, or whether Word itself is supported beyond editing files in the platform. No document management, practice management or electronic billing system is named, and MAAT LegalPaperless is a discount on a separate anonymization tool rather than a data connection.

The security page says the infrastructure runs on AWS in Ireland with confirmed EU data residency. It says customer data is segregated in a multi tenant architecture with logical isolation between organizations and users, and is encrypted at rest and in transit. The platform privacy notice of 1 July 2026 says personal data is generally not transferred outside the European Economic Area and that any transfer would rest on article 46 GDPR safeguards.

A company post of 25 February 2026 says Aptus.AI implemented Anthropic's Claude models for drafting, and article 3.3 of the general terms relies on an unnamed AI provider, yet nothing says in which region that inference runs or whether the provider is bound to EU processing. The enterprise pages promise custom configurations, but no single tenant, private cloud or on premises option is described, and no region other than Ireland is offered.

The security page says Aptus.AI holds ISO/IEC 27001:2022 certification and applies controls verified by an independent third party. A company post of 11 February 2026 announced the certification and says the company is now maintaining and improving the system. Two information security policy documents, each version 1 dated 10 July 2024 and approved by top management, define the management system's perimeter as all staff, processes, data and technologies within it, with roles for an ISMS manager and a VP of compliance and legal.

The security page adds annual external penetration testing and automated code scanning. The certifying body, certificate number, issue and expiry dates, a scope statement, and a way to request the certificate or a penetration test summary are not published. Aptus.AI has no trust center and no SOC 2 report.

A company post of 25 February 2026 on Cassazione rulings and drafting says Aptus.AI implemented Anthropic's Claude models, which it calls the best performing on the market, for document drafting. The pricing FAQ says the platform rests on proprietary AI models, with patents granted in Italy in 2023, Europe in 2024 and the United States in 2025 for the method that converts documents into its machine readable format. The December 2025 post on Next5 calls it a proprietary assistant.

Article 3.3 of the general terms and both privacy notices list an AI provider without naming it. Not published: which model serves research and which serves drafting, model versions, where inference runs, whether any other provider is used, and any commitment to notify customers when a model or provider changes. Neither privacy notice carries a subprocessor list.

On the pricing page, Standard costs 59 euros a month on the annual plan or 69 euros month to month, excluding VAT, for one user with up to 50 AI messages a month and 200 MB of storage. Professional costs 110 euros a month on the annual plan or 129 euros month to month with one user included, 49 euros a month for each extra user up to 50, unlimited AI use and 2 GB of extensible storage. The annual plan can be paid at once or monthly, and a seven day free trial needs no upfront payment.

Article 8 of the general terms refers setup and professional services fees to unpublished particular conditions. Article 13 renews automatically unless canceled ten days before term, and article 16 allows changes on 15 days' notice with a right to withdraw. The enterprise editions for large firms, in house teams and compliance functions are priced in a consultation, with no published figure, unit or tier, and nothing says what implementation adds.

The lawyers' pages split the offer by size: Standard for individual lawyers and small firms, and Professional for structured firms or teams. The enterprise pages address Italy's largest firms, in house legal departments and compliance functions in regulated sectors. The pricing FAQ lists lawyers and law firms first, then accountants and labor consultants, public administration, in house counsel and compliance professionals, and excludes private users and people without legal or compliance expertise.

The sources page names the issuing bodies: Italian ministries, courts from the Constitutional Court and the Cassazione to the regional administrative and merit courts, more than thirty Italian independent authorities, EU institutions and agencies, and financial regulators in the United States, the United Kingdom, Luxembourg, France, San Marino, Hong Kong and Singapore. No page states depth by practice area, and the benchmark covers only civil, labor and tax law. The Spanish and German editions sell the same plans without naming any Spanish or German source.