ADG 2.0 Expands AI Governance Controls From 12 to 180+

EC-Council's updated Adopt, Defend, Govern framework, ADG 2.0, replaces the original 12 minimum controls with more than 180 controls organized into 12 control families. Released on October 1, 2026, the update maps these controls to over 90 regulations, standards and frameworks, including the EU AI Act, the NIST AI Risk Management Framework and ISO/IEC 42001. The central goal is to shift AI governance from static policy documents to demonstrable operational proof, which is critical as AI agents gain the ability to use company tools and access internal data.

What changed in ADG 2.0?

The most visible change is scale. EC-Council's official framework documentation describes a library of more than 180 controls containing over 720 requirements, distributed across 12 families. The largest families include Build and Integration with 29 controls, Runtime Defense and Monitoring with 25, Oversight and Review with 23, and Evaluation and Red Teaming with 21. These sit inside three pillars: Adopt governs what may be deployed; Defend addresses security, evaluation and runtime monitoring; and Govern covers accountability, transparency and oversight.

The framework also publishes extensive crosswalks. EC-Council reports 140 mapped controls for the NIST AI RMF, 139 for the EU AI Act, 135 for ISO/IEC 42001 and 72 for MITRE ATLAS. The launch announcement notes additional mappings to MITRE ATT&CK, the OWASP Top 10 for LLM Applications and the OWASP Agentic Top 10. EC-Council has offered the framework, crosswalks and an AI Readiness Self-Assessment free to governments, regulators and standards bodies to help them apply the controls to local requirements.

Operationally, ADG 2.0 introduces four runtime checkpoints: ADMIT, DECIDE, EMIT and COMMIT, and calls for human approval at defined decision thresholds. According to the implementation details, the framework defines three autonomy tiers: Assistive, Conditional and Autonomous. Each tier specifies what a system may do, when human review is required and what evidence must be retained. That evidence includes timestamps, agent identities, permissions, material tool calls, policy checks, human approvals and overrides. EC-Council founder Jay Bavisi stated in the announcement that most AI governance today is a policy document nobody tests, and the framework is intended to replace that with evidence of actual control operation.

Why does this matter for CISOs and GRC teams?

The update arrives as enterprise AI governance is shifting from paper to proof. Gartner argued earlier in 2026 that organizations need to move beyond high-level policies to controls embedded in operational workflows, because addressing governance only after deployment makes it far harder to incorporate oversight where it is needed. ADG 2.0 fits directly into that trend by providing a control library that is meant to be tested and evidenced rather than simply referenced in a policy PDF.

For security and compliance leaders, the practical implication is that regulatory mapping can reduce duplication, but it does not establish compliance automatically. The EU AI Act is binding legislation with phased obligations. According to the European Commission, the Act entered into force on August 1, 2024, and became generally applicable on August 2, 2026. Prohibited practices and AI literacy requirements applied earlier, from February 2, 2025. Rules for certain high-risk use cases will apply from December 2, 2027, and rules for high-risk AI embedded in regulated products will apply from August 2, 2028. A framework's mapping to the Act is a compliance aid, not proof that a business meets its legal obligations. Teams still need to interpret applicability, assess jurisdiction and role, and demonstrate that their specific systems satisfy the provisions that are in force.

Similarly, NIST's AI RMF is currently being revised, and NIST published a concept note for a critical-infrastructure profile in April 2026. Enterprises using ADG 2.0 should verify that its crosswalks remain aligned with the current versions of underlying standards as they evolve.

What risks do autonomous AI agents introduce?

The urgency behind ADG 2.0 is driven by agentic AI. Cybersecurity Insiders notes that AI agents are increasingly able to retrieve internal information, invoke APIs, manipulate business applications and execute multistep workflows. Governance therefore needs to address not just whether an AI use case was approved during a review board meeting, but what the agent can access, what actions it can take, who authorized those actions, how its behavior is monitored and what evidence remains afterward.

This changes the nature of risk. A model-level failure, such as a hallucinated tool call or an exploited prompt, can become an operational, financial or data-security incident if the agent has active credentials and permissions. Responsibility is shared but not equal. Model providers oversee their own services, while enterprises remain responsible for deployment configuration, credentials, connected tools, data access and business decisions. For CISOs, that means identity, authorization and runtime security are now central to AI governance, not peripheral.

How should enterprises implement operational AI governance?

Moving from policy to proof requires lifecycle coverage. Inventory, testing, access restrictions, monitoring, incident response and evidence collection need to operate throughout an AI system's lifecycle, not just during an annual policy review. Organizations should use the official ADG 2.0 documentation to inspect underlying control definitions, map them to existing policies and decide which requirements to implement based on their specific risk profile and regulatory exposure.

Runtime enforcement is where governance becomes real. The ADMIT, DECIDE, EMIT and COMMIT checkpoints should be enforced technically wherever possible, with human approval gates at defined thresholds for Conditional and Autonomous tiers. Evidence must be collected as the system operates rather than reconstructed before an audit. Security teams should ensure that logs capture not only model inputs and outputs but also agent identity, permission state, material tool calls, policy checks and human overrides.

For teams managing EU AI Act compliance, the framework can help structure control owners and evidence repositories, but each control must still be traced to the legal obligations that apply to the organization's specific systems. A centralized AI governance dashboard can help track control status, owner assignments and evidence gaps across multiple frameworks, though tooling alone does not satisfy the requirement for human judgment and intervention. Tools that provide runtime visibility into employee AI use across browsers, desktop apps and APIs, including Walled AI, can contribute to the evidence base ADG 2.0 expects, but the framework itself remains vendor-neutral.

What are the open questions and limitations?

Direct independent reporting on ADG 2.0 is limited. The strongest sources available are EC-Council's own announcement and framework documentation, syndicated press releases and a single detailed trade publication. That does not invalidate the controls, but it means security leaders should treat the framework as one reference among many and validate its fitness for their environment rather than accepting coverage claims at face value.

There is also a timing consideration. Because the EU AI Act and NIST AI RMF are still evolving, crosswalks can become outdated. Compliance teams should schedule periodic reviews of mapped controls against the authoritative sources rather than treating the crosswalk as a one-time alignment exercise. Finally, while the framework and self-assessment are free to access, implementing 180-plus controls requires staffing, tooling and cultural change that many organizations have not yet budgeted for.

What is the broader context for AI governance?

ADG 2.0 is part of a wider industry recognition that AI governance must be operational. The question for enterprises is no longer whether a use case was approved in principle, but whether the organization can prove what the agent did, who authorized it and how it was stopped if something went wrong. As agentic AI moves from experiment to production, frameworks that emphasize testable controls, named owners and runtime evidence will become baseline expectations for auditors, regulators and boards. The organizations that begin building that evidence now will be the ones that can demonstrate compliance when it is required, not just when it is convenient.