Collibra launches Maestro agents, Live Map, and Guardian runtime controls

Collibra announced Collibra Maestro (no-code agents with governance built in), Live Map (a context graph over curated unstructured documents for agentic retrieval), and Guardian Agents with Agent Contracts for runtime control of third-party agents — and said Maestro Assistant is the first Data and AI governance company agent to achieve AIUC-1 certification.

SOFTWARE desk — companies are drowning in agents they didn’t build; Collibra is selling both a factory for governed agents and a runtime leash for the rest of the fleet.

What the three names are, in the release’s order. Collibra Maestro lets governance teams build their own no-code agents, with governance built in from the start, to automate workloads. The release says that turns weeks of manual work into minutes. Weeks, and minutes, are the company’s claim. This desk did not time a task. Live Map gives every agent a context graph of how a company’s knowledge connects, so an agent can retrieve from curated unstructured documents without rebuilding that map on every question. Guardian Agents with Agent Contracts extend that governance to the much larger fleet of AI agents built outside Collibra. An AI agent, here, is software that can take a series of steps, not only answer one question. No-code means a person sets the agent up in a visual builder instead of writing a program. A fleet, here, is the set of agents a company already runs. These lines are the release’s.

What Maestro is made of. The release calls it an integrated suite of AI-powered capabilities, built on Collibra’s enterprise Knowledge Graph. A knowledge graph, here, is Collibra’s map of governed data: the assets, and the rules about them. The suite is meant to simplify automation of governance operations and to speed trusted context to every asset, every AI model, and every agent. It brings Collibra’s existing native AI utilities together under one name, and it adds Maestro Studio. In Studio, a team configures and manages its own agents. Those agents answer questions and perform automated governance tasks inside Collibra. A visual builder lets a user set an agent’s persona, its permissions, and its tool access, and ground it in certified knowledge sources. A persona is the voice and the job the agent is given. A permission is what it may see or do. A certified knowledge source is a source the governance team has already marked as trusted. Out-of-the-box templates jumpstart agent creation. Jumpstart is the release’s word. Once an agent is published, end users talk to it through Maestro Assistant, a chat box, in ordinary language. These lines are the release’s. This desk did not open Studio.

Live Map, and whose product it is. The release says it comes from Deasy Labs by Collibra, which specializes in unstructured data. Unstructured, here, means documents and files that are not already rows in a table. Live Map gives agents a context graph of which entities exist across a company’s documents, how they relate, and how those relationships change. An entity is a thing the documents name, such as a customer, a product, or a policy. A context graph is that map of things and links. The release says that without it, an agent has to reconstruct those relationships for every question. Live Map makes the map available. The release says that enables more performant agentic retrieval from curated unstructured content. Curated means someone already chose which documents are in the set. Agentic retrieval means the agent goes and fetches what it needs, instead of being handed one fixed file. Performant is the release’s word. These lines are the release’s. This desk did not query a document set.

The controls for agents Collibra did not build. Agent Contracts are layered, machine-readable controls that scale across the agent fleet. Machine-readable means software can apply the rule, not only a person reading a policy. The release says the controls include global rules and specialized requirements based on function, risk classification, or operating context. A global rule applies across the fleet. A specialized requirement changes with the job, the risk label, or where the agent is running. Guardian Agents carry those controls into runtime. Runtime means while the agent is acting, not only on the day someone designed it. They continuously supervise one agent’s behavior and can step in when it moves outside the boundaries it was given. The release’s three steps are flagging an issue, escalating it, and blocking an unauthorized action. Flag means mark it. Escalate means send it to a person. Block means stop the action. These lines are the release’s. This desk did not watch an agent get stopped.

The survey the release cites, and whose numbers they are. The 2026 Hallucination Tax Report is a survey The Harris Poll ran for Collibra. The release says it was released on September 16. It found that 76 percent of data and AI decision-makers have hit critical roadblocks moving agents from pilot into production over the past 12 months. A pilot is a trial. Production means the agent is doing real work. Seventy-six percent is about three of those decision-makers in four, if the rate holds. The same release says nearly nine in ten decision-makers, printed as 87 percent, say their teams regularly re-verify that the context their agents rely on is still accurate and current. Just over half, printed as 51 percent, report spending significant staff hours manually reviewing and correcting autonomous agent outputs before they go live. That share rises to 64 percent at organizations with $100 million or more in revenue. Autonomous, here, means the agent takes steps on its own. These figures are the release’s account of that Harris Poll survey. This desk did not re-read the poll. The release points readers to collibra.com for the data and does not print a report URL. A survey claim is not a count this desk made.

The certification, and the three ways the page says first. The subhead says Collibra becomes the first Data and AI governance platform company to achieve AIUC-1 certification. The body says Collibra achieved AIUC-1 certification for Maestro Assistant, Collibra’s AI-powered assistant, and that this makes Collibra the first Data and AI governance company to have one of its agents certified against AIUC-1. Rajiv Dattani, identified as co-founder of the Artificial Intelligence Underwriting Company, says that by certifying Maestro Assistant, Collibra becomes the first Data and AI governance company to achieve AIUC-1 certification for one of its agents. Platform company is the subhead’s extra pair of words. The body and the quote do not print platform. File each wording with its sentence. AIUC-1 is the standard the release names. It says the certification checks Maestro Assistant against requirements for AI security, safety, and reliability. The areas it names include data and privacy, security, unauthorized actions, monitoring, and human oversight. Including means the release does not call that list complete. An underwriting checklist, in ordinary speech, is the list a reviewer uses before standing behind a system. First is the company’s claim, and Dattani’s. This desk did not audit a certificate.

The assessment that was already in the product. Since May, the release says, AI Command Center has included an out-of-the-box AIUC-1 assessment. Out of the box means it ships with the product. The assessment is there so an organization can score its own AI against the AIUC-1 requirements, find the gaps, and take steps toward alignment. May is earlier than this 23 Sep announcement. The assessment is a tool for a customer’s own AI. Maestro Assistant’s certification is Collibra’s own agent passing that standard. Those are two sentences. These lines are the release’s.

The two quotes, and only the release. Felix Van de Maele, chief executive, said governance teams are carrying a growing manual workload, which is exactly why Maestro automates governance operations directly within Collibra. He said the much bigger challenge is everything else. Agents a team bought, built on other platforms, or stood up without governance in mind are all subject to the hallucination tax. He said Live Map gives every one of those agents the context it would otherwise have to reconstruct on its own. He said Agent Contracts and Guardian Agents let organizations set the rules of engagement for that entire fleet and enforce them at runtime. He said that is how the company thinks organizations will scale from a handful of agents to a fleet of thousands. A handful, and thousands, are his scale on the page. They are not a customer count. Rajiv Dattani said that as AI agents take more actions across enterprise systems, organizations need clear and rigorous standards for security, safety, and oversight. He said certifying Maestro Assistant sets an important benchmark for the industry and demonstrates its commitment to scaling agentic AI responsibly. Its, on that sentence, is the release’s word. Those are their sentences. This desk did not interview them.

When a customer can use each piece, in the release’s tense. Customers can access Collibra Maestro today, including Public Preview availability for Maestro Studio and Maestro Assistant. Public Preview is the release’s word for those two pieces. The page does not print the words general availability. Agent Contracts and Guardian Agents will be available through the AI Command Center in October. Attendees can see both in action at the Collibra AI Summit in New York on October 6 and in London on October 21. The page does not print a year beside October 6 or October 21. Live Map is opening this quarter to a small group of design partners. This quarter is the release’s phrase. A design partner is a customer trying the product with the company. It is not the full customer list. Today, Public Preview, October, October 6, October 21, and this quarter are the release’s clocks. This desk did not sign in.

What the about box says the company is. Collibra says it is the enterprise AI control plane, built over eighteen years of ontology engineering. Ontology, here, is the set of names and relationships a company uses for its data. It says the product governs context and control across any data, any model, and any agent, for AI leaders who need certainty at the speed of AI. It says it is a five-time Forrester Wave Leader and a two-time Gartner Magic Quadrant Leader. Those are analyst rankings the about box claims. This desk did not re-read the reports. It says it serves more than 700 customers, including 78 of the Fortune 500, and governs more than 2 billion assets. More than 700, 78, and more than 2 billion are the about box’s figures. They describe the company. They are not a count of Maestro users. The site it names is collibra.com. These lines are the about box.

Plain English for the rest of the card. Hallucination tax = the extra cost and the rework when an AI agent invents a wrong answer. The release’s version is a hidden cost: higher cost, more errors, a hit to productivity, and less trust. AIUC-1 = the Artificial Intelligence Underwriting Company’s checklist for agent security, safety, and reliability. Maestro Studio = the visual builder where a governance team configures a no-code agent. Maestro Assistant = the chat where a person talks to an agent that has been published. A knowledge graph = Collibra’s map of governed data. A context graph = the map of entities and relationships across documents. Live Map = that context graph, from Deasy Labs by Collibra. An Agent Contract = a machine-readable rule sheet: global rules, plus rules that change with the job, the risk, or the setting. A Guardian Agent = a supervisor that can flag, escalate, or block an agent at runtime. Runtime = while the agent is acting. Public Preview = Studio and Assistant, available to try with Maestro today. October = when Contracts and Guardian Agents arrive in the AI Command Center. October 6 and October 21 = the New York and London summit dates the release prints, with no year beside them. This quarter = Live Map, for a small group of design partners. 9:00 a.m. Eastern = the wire’s stamp. 76 percent, 87 percent, 51 percent, and 64 percent = the release’s Harris Poll figures. $100 million = the revenue line where the 64 percent figure applies. September 16 = the day the release says the survey came out. 700, 78, and 2 billion = the about box. This filing is the 23 Sep announcement.

PRIMARY here: Collibra’s 23 Sep 2026 PR Newswire release, stamped Sep 23, 2026, 09:00 ET and datelined New York and Brussels — Tier A PRIMARY, the company’s own announcement. The three capabilities, the weeks-into-minutes claim, the Knowledge Graph, Maestro Studio, and Maestro Assistant description, the Deasy Labs by Collibra line, the context-graph sentence, the Agent Contracts and Guardian Agents description, the Harris Poll 76 percent, 87 percent, 51 percent, and 64 percent figures, the September 16 report date, the three first sentences, the AIUC-1 areas, the May AI Command Center assessment, the Van de Maele and Dattani quotes, the handful-to-thousands line, today’s access with Public Preview for Studio and Assistant, the October availability through AI Command Center, the October 6 New York and October 21 London summit dates, the design-partner line for Live Map, and the about-box figures are that release’s. NOT claimed: that this desk opened Studio, signed a certificate, or re-read the Harris Poll, a price, a named Maestro customer, a report URL past collibra.com, that the subhead’s platform company and the body’s governance company are one printed phrase, that the May assessment is the same event as the Maestro Assistant certification, that 700 customers are Maestro users, the words general availability, a year printed beside October 6 or October 21, a stock tip, or investment advice. Distinct from the already-filed brain-company-memory, zerodrift-anchor-3, wisdomai-live-apps, and deepmind-private-ai-memory.