# Work with AI agents | Product discovery scaffold - Kind: Detail - URL: /it-it/work-with-ai-agents - Canonical: (none) - HTML lang: it-IT - H1: Work with AI agents - Description: Exploration scaffold for supervised AI-agent workflows and clear human control. - Structured data: BreadcrumbList, ListItem, WebPage - Internal links: 77 - Image coverage: 3/11 images with alt text - Summary: Work with AI agents | Product discovery scaffold. Exploration scaffold for supervised AI-agent workflows and clear human control.. Work with AI agents - Related links: it-it/product-connect-and-prepare-data, it-it/product-collect-and-keep-evidence-current, it-it/product-report-with-confidence, it-it/work-with-ai-agents, it-it/automation - Outline: Section: Articoli più recenti Canadian report maps security work in small businesses A review of public data looks at security st | Section: Scelti dalla redazione Unauthorized radio signals stopped trains in Poland in 2023 In August 2023, unauthorized radio tr | Section: Esplora Capire le minacce e i cambiamenti 9 Prendere decisioni sulla sicurezza 1 Prepararsi, rispondere e ripartire 2 Pr ## Section Articoli più recenti Canadian report maps security work in small businesses A review of public data looks at security staffing, written policies and updates among Canadian businesses with 10 to 49 employees. INTERCEPT hosts free cybersecurity webinar for small teams The 18 November online event will cover phishing, multi-factor authentication and a cybersecurity learning platform. A third-party software vulnerability led to a data breach at Frontline Education Frontline has begun notifying people affected by an August incident involving two school administration products. The impact varies between districts and individuals. ## Section Scelti dalla redazione Unauthorized radio signals stopped trains in Poland in 2023 In August 2023, unauthorized radio transmissions triggered Poland’s railway stop function. Reporting said the receiver did not verify who sent the command; no one was injured. A cyberattack disrupted refrigeration at hundreds of French supermarkets An attack on 12 October 2026 changed refrigeration settings and led stores to discard fresh and frozen food. CERT-FR named several affected controller families, warned that activity might continue and did not identify how access began. A secret is not rotated until the old credential stops working Railway’s preliminary report says some environment variables in at most 1.85% of workspaces may have been compromised, but it could not identify which specific variables were affected. Prove a rotation is complete by checking each consumer and confirming the old credential is revoked. ## Section Esplora Capire le minacce e i cambiamenti 9 Prendere decisioni sulla sicurezza 1 Prepararsi, rispondere e ripartire 2 Proteggere l’attività ogni giorno 1 ## Section Argomenti ## LearnEditorialMobileSection Articoli più recenti Canadian report maps security work in small businesses A review of public data looks at security staffing, written policies and updates among Canadian businesses with 10 to 49 employees. INTERCEPT hosts free cybersecurity webinar for small teams The 18 November online event will cover phishing, multi-factor authentication and a cybersecurity learning platform. A third-party software vulnerability led to a data breach at Frontline Education Frontline has begun notifying people affected by an August incident involving two school administration products. The impact varies between districts and individuals. Vedi tutti gli articoli → ## LearnEditorialMobileSection Guide pratiche Rispondi ai questionari di sicurezza senza caos Un metodo passo passo per rispondere più velocemente senza improvvisare ogni risposta da zero. Vedi tutte le guide → ## LearnEditorialMobileSection Scelti dalla redazione Unauthorized radio signals stopped trains in Poland in 2023 In August 2023, unauthorized radio transmissions triggered Poland’s railway stop function. Reporting said the receiver did not verify who sent the command; no one was injured. A cyberattack disrupted refrigeration at hundreds of French supermarkets An attack on 12 October 2026 changed refrigeration settings and led stores to discard fresh and frozen food. CERT-FR named several affected controller families, warned that activity might continue and did not identify how access began. A secret is not rotated until the old credential stops working Railway’s preliminary report says some environment variables in at most 1.85% of workspaces may have been compromised, but it could not identify which specific variables were affected. Prove a rotation is complete by checking each consumer and confirming the old credential is revoked. ## LearnEditorialMobileSection Esplora Categorie degli articoli Capire le minacce e i cambiamenti 9 Prendere decisioni sulla sicurezza 1 Prepararsi, rispondere e ripartire 2 Proteggere l’attività ogni giorno 1 Argomenti ## CmsPage Work with AI agents Exploration scaffold for supervised AI-agent workflows and clear human control. Visualizzazione del contenuto originale in `en-gb` in attesa che questa lingua sia tradotta. Exploration scaffold. This page captures the intended direction for Aeptus to support teams working with AI agents in security management, with people retaining control. It does not claim that agent access, protocols, or autonomous actions are available in the released product. Questions to answer Which tasks should an AI agent be able to read, draft, recommend, or carry out? Which actions must always wait for a person’s review or approval? How does an agent receive only the workspace data and permissions it needs? How are agent actions attributed, recorded, reviewed, corrected, or reversed? Which agent interfaces or protocols are part of the product direction, and which are implemented? How are untrusted inputs, sensitive information, and failed or interrupted actions handled? Product and code information needed The agent-facing API, tool, or protocol contract and its supported operations. Authentication, workspace boundaries, least-privilege scopes, and read/write action matrix. Human approval gates, audit records, provenance, cancellation, and recovery behavior. Security controls for untrusted content and sensitive data. Feature flags, release notes, and code evidence distinguishing plans from shipped behavior. Public-language guardrails Describe the intended supervised workflow only. Do not claim that Aeptus is agent-enabled or acts autonomously until released product behavior has been verified.