VerifyWise 2.4: govern third-party AI, AI apps, and agents

Most of the AI risk in your company is not in the models you trained. It is in the tools your teams already use, the third-party apps they signed up for last week, and the agents now writing code and moving files on their own.

The numbers back this up. In IBM's Cost of a Data Breach Report 2025, 63% of breached organizations had no AI governance policy in place to manage AI or rein in shadow AI, and 97% of those that suffered an AI-related security incident lacked proper AI access controls. Gartner, meanwhile, predicts that over 40% of agentic AI projects will be canceled by the end of 2027, often because of unclear value and weak risk controls. AI spreads through a company faster than governance catches up, and that gap is where most of the risk sits.

VerifyWise 2.4 is built to close that gap. This release adds three new modules and a set of supporting features that bring the same governance discipline to the wider surface of AI your organization runs on in practice: the apps it buys, and the agents it sets loose. Here is what shipped and how to use it.

What is new in 2.4

2.4 is a major release with three new modules at its center. The AI Trust Index grades and tracks third-party AI apps. The AI Apps inventory governs the AI tools your own teams use. Agent Control governs what AI agents are allowed to do. Around those, the release adds custom fields, a deadline warning banner, Microsoft Entra ID single sign-on, and Spanish coverage across the application.

All three modules share one idea. For years, governance covered the models a company owned while the risk spread to AI it only used. 2.4 extends the same review, approval, and audit discipline to AI you did not build but still depend on.

AI Trust Index: grade the AI apps before you trust them

Your teams adopt third-party AI apps faster than any review process can keep up. Each one handles your data under its own terms, trains on it or does not, retains it or deletes it, and most people signing up never read the policy. The AI Trust Index gives you a way to judge those apps before you trust them.

The Index is a curated catalog of AI applications, each carrying a letter-grade trust score with a per-domain breakdown. You open an app's page to see its grade, the verdict behind it, and the watch-outs worth knowing before you let it near company data. You can track the apps you care about and get a weekly digest email of what changed, so a quietly updated data policy does not slip past you.

This grew out of our own research. We scored the privacy policies and terms of hundreds of widely used AI apps against one consistent rubric, and the results were not reassuring. We wrote that up in our analysis of 205 AI apps graded on data transparency, where only 23% earned an A or B. The Index turns that one-time study into a living reference you can check whenever a new tool lands on your desk. You can browse it now on the AI Trust Index.

AI Apps inventory: turn shadow AI into governed AI

Knowing how trustworthy an app is solves half the problem. The other half is knowing which apps your organization runs day to day, and doing something about the ones nobody approved. Unsanctioned AI tools are common, and they are exactly the blind spot the IBM breach data points to.

The AI Apps inventory is a register of the AI applications your teams run. Each app gets a detail page. An approval center routes new apps through review and sign-off. You can map an app's policy and model dependencies, so you can see which policy covers it and which underlying model it leans on. Risk assessment templates let you evaluate each app on a consistent basis rather than improvising every time.

Promotion is what ties this to the rest of the platform. When a discovered shadow AI tool turns up, you can promote it straight into a governed app, with its dependencies and risk assessment attached. A tool you did not know about yesterday becomes a tracked, assessed entry in your register.

If shadow AI is a live concern for you, our guide on shadow AI detection covers how to find unsanctioned usage in the first place. The AI Apps inventory is where that discovery turns into governance.

Agent Control: govern what AI agents do

An AI agent carries more risk than a chat app. It calls tools, writes files, and chains actions together without a human reviewing each step. OWASP names this in its Top 10 for LLM Applications as "Excessive Agency," the risk of giving a model the autonomy to act without adequate checks. The Gartner number on agentic projects failing for weak risk controls is the same problem measured in budgets.

Agent Control governs agent behavior at the point of action. It gates two things that matter most: tool calls and file writes. An agent that wants to use a tool or write to a file passes through a control gate first. Approved actions run. Actions that need oversight get held and flagged. Either way, the platform captures the tool results and builds an events timeline, so you have a record of what the agent did and why.

The new Runs page is where this comes together. Run correlation joins an agent's model calls and tool calls into a single run, so you can follow one agent's behavior end to end instead of reading scattered logs. Agent Control works with Claude Code and Cursor, the agentic coding tools your engineers are most likely already using.

An autonomous agent then gets the same approval step and audit trail you would expect for any other system touching your data.

The supporting features

The three modules carry the headline, but the smaller features are what change how governance feels day to day. 2.4 adds several of them.

Custom fields let you define your own fields to capture the data your organization needs, rather than bending your records to fit a fixed schema. If your risk register or app inventory needs a field the platform does not ship by default, you add it.

The deadline warning banner surfaces upcoming deadlines across the dashboard and tasks. You set the thresholds, and each user can snooze a warning that does not apply to them. Missed compliance deadlines are a quiet, common failure in governance programs, and a visible banner is a simple guard against them.

Microsoft Entra ID single sign-on lets people sign in with your organization's Azure AD identity. That means central identity control for IT, one less password for everyone, and every sign-in landing in the audit trail. We first shipped enterprise SSO in VerifyWise 1.5, and 2.4 extends that work.

The release also adds Spanish across the application, alongside the German and French coverage already in place, plus a round of design system updates: a unified chip component, icon-only and small button sizes, table skeleton loaders on more pages, and consistent empty states.

What this covers for compliance

The EU AI Act entered into force on 1 August 2024 and becomes fully applicable on 2 August 2026, with penalties reaching 35 million euros or 7% of global annual turnover for the most serious violations. It expects a maintained inventory of your AI, documented risk assessment, and an audit trail. An AI apps register and an agent run history produce those records as a side effect of using them.

If you are mapping your obligations, our explainer on governing high-risk AI under the EU AI Act and the comparison of the EU AI Act and ISO 42001 are good places to start. VerifyWise supports the EU AI Act, ISO 42001, and ISO 27001 as built-in frameworks, so the work you do in these new modules feeds the same compliance picture as the rest of the platform.

The supporting research says the same thing. Stanford HAI's 2025 AI Index reports rising AI adoption while responsible-AI practices lag behind, and Cisco's 2026 Data and Privacy Benchmark finds organizations taking on AI responsibilities they are not ready for. Most companies adopt AI well before they govern it.

How to get started

If you already run VerifyWise, 2.4 is available now and the three new modules appear in your sidebar. A good first move is to open the AI Apps inventory and register the AI tools you know your teams use, then check each against the AI Trust Index. If your engineers run agentic coding tools, connect Agent Control to one of them and watch a few runs land on the Runs page.

If you are new to VerifyWise and weighing whether to build governance tooling yourself or adopt a platform, our write-up on build versus buy for AI governance lays out the trade-offs honestly.

VerifyWise is source-available, and you can self-host it or run it as a managed deployment. The full v2.4 release notes are on GitHub. To see how the new modules fit your own AI footprint, book a demo and we will walk through it with your tools in mind.