Summary
- Vanta has signed a definitive agreement to acquire Pillar Security, bringing AI agent monitoring, testing, and enforcement capabilities into its governance platform.
- AI inventory and runtime monitoring are planned for January 2027, followed by adversarial testing in spring and enforcement in summer.
- The planned integration would connect policy requirements with operational controls, although the product's effectiveness and coverage remain to be demonstrated.
Vanta has agreed to acquire AI security specialist Pillar Security, aiming to connect corporate governance policies with controls that can monitor autonomous agents and prevent unauthorised actions while they operate.
The acquisition, announced on 9 November, would add runtime protection and adversarial testing to Vanta’s existing trust and compliance platform. The company plans to integrate Pillar’s technology with its Trust Graph, which draws information from more than 400 integrations to connect business systems, controls, responsibilities, and risks.
The transaction has been agreed but has not been confirmed as completed, and its financial terms were not disclosed. Vanta intends to introduce the combined capabilities progressively through 2027, with AI inventory and monitoring planned for January and runtime enforcement scheduled for the summer.
Pillar Security, founded in 2023 by Ziv Karliner and Dor Sarig, develops technology for discovering AI agents, examining their security weaknesses, monitoring activity, and applying controls to their behaviour. Its acquisition would extend Vanta’s offering beyond the documentation and assessment of governance requirements.
Autonomous agents introduce different risks
Traditional enterprise applications generally perform actions through defined workflows and access permissions. Autonomous AI agents can introduce additional complexity because they determine sequences of steps, interact with connected tools, and make decisions about how to complete assigned tasks.
An agent authorised to analyse business information might retrieve documents, update records, send messages, or invoke software services. Each individual capability may be legitimate, while an unexpected combination of actions can create risks that were not fully anticipated when permissions were granted.
Those risks can arise through mistakes in an agent’s reasoning, poorly defined instructions, compromised information sources, or deliberate attempts to manipulate its behaviour.
A system permitted to read sensitive financial information, for example, should not automatically be able to transfer that information through an unapproved communications channel. Similarly, an agent authorised to modify a particular set of records should not gain unrestricted access to unrelated business systems.
Organisations have traditionally addressed these concerns through access management, security policies, reviews, and technical monitoring. Those controls remain necessary, but the autonomy of AI agents creates a greater requirement to understand what systems are doing between formal assessments.
Vanta wants to connect governance documentation with evidence collected during actual operations, making it possible to identify when an agent attempts an action outside its approved responsibilities.
Governance moves closer to operational controls
The planned integration will use Vanta’s Trust Graph to provide organisational context for events identified through Pillar Security’s monitoring technology.
A conventional security alert may show that software attempted to access a particular system. Deciding whether that action was appropriate can also require knowledge of the software’s owner, its approved purpose, the data involved, and the restrictions established by the organisation.
Vanta’s proposed architecture would connect that information with runtime protection. When an agent attempts an unauthorised action, an enforcement mechanism could prevent the activity and record evidence of the intervention.
The company intends to use those records to support investigation and compliance processes, reducing the gap between a policy specifying what an agent should do and evidence of what it actually attempted.
Such controls could be useful to organisations operating AI agents across financial systems, customer databases, development environments, and other applications where automated actions have operational consequences.
However, preventing an action depends on where enforcement takes place. A security platform can intervene only when the relevant activity passes through a mechanism capable of inspecting and controlling it.
Coverage will therefore depend on the integrations supported, the permissions available to the security system, and whether agents can perform activities through tools or communications paths outside the protected environment.
Organisations must also define which activities should be blocked, which should generate alerts, and which require human approval. Excessively restrictive controls could interrupt legitimate business operations, while incomplete rules may leave sensitive activities insufficiently protected.
Inventory comes before enforcement
Vanta has outlined a staged integration programme, beginning with AI inventory in January 2027. The capability is intended to identify applications, models, agents, and related services, including Model Context Protocol servers.
It would associate those components with their owners, approved use cases, data access, vendors, and relevant security controls. Establishing an accurate inventory is an important prerequisite for applying consistent governance because organisations cannot reliably manage systems whose deployment and permissions they do not understand.
Runtime monitoring and detection are scheduled for the same month. These tools would identify suspicious behaviour or potential policy violations, providing information that security teams could use to prioritise investigations.
Adversarial testing is planned for spring 2027. The capability would allow organisations to examine how agents respond to attempts to exploit weaknesses in their configuration or behaviour, including whether they can be induced to perform actions outside their approved boundaries.
Testing may identify vulnerabilities before an attacker can exploit them, although exercises must be appropriately authorised and controlled to avoid interfering with operational systems.
Runtime enforcement is expected in summer 2027. This would introduce mechanisms intended to block unauthorised activity rather than merely detect it after the event.
The timetable represents Vanta’s planned product development rather than a commitment that all capabilities are available today. The company has not published independent measurements demonstrating the effectiveness of the proposed integration across different enterprise agent environments.
Adoption is moving ahead of oversight
Vanta’s earlier research found that 65% of surveyed organisations believed their use of agentic AI was advancing faster than their understanding of it. Only 48% reported having a framework for granting or limiting autonomy to AI systems.
Those findings provide context for the acquisition, although they are vendor-reported survey results rather than evidence that every organisation faces the same level of exposure.
As businesses introduce autonomous applications, governance teams must consider who authorises access, how permitted activities are defined, and how unexpected behaviour can be investigated. These requirements extend beyond the technology used to generate decisions and include the services through which agents interact with the organisation.
For European businesses, runtime evidence may help support existing governance, cybersecurity, and data protection responsibilities, particularly where automated systems interact with sensitive information or business-critical applications.
However, installing a monitoring or enforcement product does not itself establish regulatory compliance. Organisations remain responsible for assessing their legal obligations, managing data access, and ensuring that security arrangements operate effectively within their own environments.
Vanta says it now serves more than 20,000 customers, providing an established commercial base for the planned AI security offering. Its existing compliance and risk management products also give the company access to organisations already collecting evidence about security controls and third-party dependencies.
Combining that information with runtime protection could expand the role of governance platforms, which have traditionally focused on policies, assessments, and audit preparation rather than direct intervention in application behaviour.
The commercial outcome will depend on whether the integrated products provide sufficient coverage and reliability for organisations to use them in operational environments. Security teams will need evidence that interventions are accurate, that legitimate processes are not unnecessarily disrupted, and that the system maintains clear records of the actions it permits or blocks.
Vanta expects its paid AI Governance offering to become available in January 2027, with further Pillar Security capabilities introduced during the year. The planned enforcement functionality remains several months away, leaving its practical performance to be assessed as customers begin deploying the combined technology.












