AI Guide

EU AI Pact: The voluntary scheme ahead of the EU AI Act

The EU AI Pact is the European Commission's voluntary pre-compliance scheme that invites companies to commit to AI governance practices before the EU AI Act's binding obligations fully apply. It is not the AI Act itself and not an enforcement body, but a bridge of preparation built around three voluntary pledges. This article explains who runs the Pact, what signatories commit to, and how it differs from the Act and the EU AI Office.

Key Facts
  • The European Commission launched the AI Pact on 25 September 2024, with more than 100 initial signatories including Amazon, Google, Microsoft, and OpenAI.
  • By December 2025 the Commission reported roughly 3,265 companies and organizations engaged, with more than 230 having made formal voluntary pledges.
  • Signatories commit to at least three core actions: an AI governance strategy, mapping of high-risk AI systems, and promoting AI literacy among staff.
  • The Pact is not legally binding; declining to sign or failing to follow through carries no fine, unlike noncompliance with the EU AI Act itself.
  • Enrollment stays open on a rolling basis until the EU AI Act's obligations fully apply, with high-risk provisions phasing in through 2027.

Definition: EU AI Pact

The EU AI Pact is a voluntary initiative launched by the European Commission that invites organizations to commit publicly to specific AI governance practices before the EU AI Act’s binding obligations take full effect.

Core characteristics of EU AI Pact

The Pact runs on two pillars: an open knowledge-exchange network for any interested organization, and a second track where companies make formal, trackable pledges.

  • Open to any organization developing or deploying AI in the EU, regardless of size or sector
  • Signatories commit to at least three core actions, with further pledges optional
  • Entirely voluntary: no fee, no legal force, no penalty for non-adherence
  • Administered by the European Commission, separate from the Act’s enforcement bodies

EU AI Pact vs. EU AI Act

The distinction matters for compliance planning. The Act is binding law, Regulation (EU) 2024/1689, with risk tiers and fines up to 35 million euros or 7% of global turnover for the most serious violations. The Pact is a voluntary pre-compliance gesture: signing it shows intent and builds internal readiness, but it satisfies no Article of the Act on its own. A company can ignore the Pact entirely and still be fully compliant once deadlines arrive, just as a company can sign the Pact and still fail a later conformity assessment. The Pact is preparation, not compliance.

Importance of EU AI Pact in enterprise AI

For Mittelstand companies, the Pact is a low-risk way to build AI governance habits before high-risk AI system obligations become mandatory. Adoption has scaled fast: from just over 100 signatories at launch in September 2024 to roughly 3,265 engaged organizations by December 2025, according to the Commission’s own progress reporting.

Methods and procedures for EU AI Pact

Joining follows a simple, self-paced process built around three mandatory commitments.

Pledge 1: AI governance strategy

Signatories adopt an internal AI governance strategy that supports responsible AI adoption and lays groundwork for future compliance.

  • Designate internal ownership for AI governance decisions
  • Document how AI projects are approved, reviewed, and retired
  • Align the strategy with existing AI governance frameworks already in use

Pledge 2: Mapping high-risk systems

Organizations map, to the extent feasible, which AI systems would likely fall under the Act’s high-risk categories, giving them a head start on the inventory work the Act will later require formally.

Pledge 3: Promoting AI literacy

Signatories promote AI literacy among staff and anyone acting on the organization’s behalf with AI systems, tailored to each person’s role and exposure to AI tools.

Important KPIs for EU AI Pact

Because the Pact carries no legal reporting duty, organizations track their own internal readiness indicators.

Governance readiness indicators

  • AI governance strategy: drafted and approved
  • High-risk system mapping: percentage of AI tools inventoried
  • Staff literacy training: percentage of relevant staff trained
  • Pledge coverage: additional voluntary pledges adopted beyond the three core ones

Adoption momentum

The Commission’s one-year update reported over 3,265 participating organizations by December 2025, roughly 63% EU-based and 23% US-based, showing the Pact has become an international signal rather than an EU-only exercise.

Follow-through quality

Beyond sign-up counts, the more meaningful KPI is whether a signatory’s internal processes actually reflect its pledges once a real AI audit or regulator inquiry arrives, since the Pact itself verifies nothing.

Risk factors and controls for EU AI Pact

Treating the Pact as a substitute for real compliance work creates its own risks.

Mistaking the Pact for compliance

The biggest risk is a board believing that signing satisfies EU AI Act obligations.

  • Confusing a voluntary pledge with a binding legal requirement
  • Under-resourcing the real high-risk inventory because “we already mapped it for the Pact”
  • Treating literacy promotion as a one-off communication rather than an ongoing duty

Reputational exposure without delivery

Signing publicly and then visibly failing to follow through carries reputational risk even though it carries no fine, since the Commission and press track signatory lists and progress reports.

Confusing the Pact with the EU AI Office

Some companies assume the Pact is run by the EU AI Office, the body enforcing general-purpose AI model rules under the Act. The Office has no role in collecting or policing Pact pledges; the Pact is a separate Commission initiative open to any organization, not only GPAI providers, and carries none of the Office’s binding enforcement powers.

Practical example

A 140-employee specialty chemicals distributor in North Rhine-Westphalia signed the AI Pact in early 2025 so its board could take a visible first step on AI governance before the Act’s deadlines made one mandatory. The compliance lead used the three pledges as a structured starting checklist rather than a bureaucratic exercise. Within four months the company had a one-page governance charter, a risk-tiered inventory of its twelve internal AI tools, and a short training module assigned to the 35 staff who regularly use AI.

  • Board-approved AI governance charter reused later as the basis for formal compliance documentation
  • Risk-tier inventory that fed directly into the company’s later EU AI Act classification work
  • Role-specific AI literacy training distinguishing casual users from staff building automations
  • A single internal point of contact for any future regulator inquiry

Current developments and effects

The Pact continues to evolve as the Act’s own deadlines approach.

Sustained signatory growth

Adoption has not plateaued since launch.

  • Growth from roughly 100 signatories in September 2024 to over 3,265 by December 2025
  • Weekly additions reported directly by the Commission
  • Expanding geographic spread beyond the EU and United States

Overlap with Digital Omnibus discussions

The Digital Omnibus proposal, still under negotiation in 2026, has pushed parts of the high-risk compliance calendar later without changing the Pact’s voluntary, rolling-enrollment nature. A longer runway to enforcement makes early Pact participation more useful as practice time.

Scrutiny of pledge follow-through

Commentators and some member-state authorities are starting to ask whether pledges translate into real internal change, a question that will sharpen as more signatories near the Act’s actual enforcement dates.

Conclusion

The EU AI Pact is best understood as a voluntary warm-up, not a shortcut around the EU AI Act’s binding requirements. Its three pledges, an AI governance strategy, a high-risk system mapping, and staff AI literacy promotion, give organizations a structured way to start governance work early. It is administered by the Commission but enforced by nobody, distinct from both the Act itself and the EU AI Office’s binding role. Companies that treat it as genuine preparation, not a public relations checkbox, tend to arrive at their first real conformity assessment considerably better prepared.

Frequently Asked Questions

Is signing the EU AI Pact legally required?

No. The Pact is entirely voluntary. There is no fine for not signing, and signing creates no new legal obligation beyond what the EU AI Act itself already requires once its deadlines apply.

How is the EU AI Pact different from the EU AI Act?

The Act is binding law with risk tiers, documentation duties, and fines. The Pact is a voluntary pledge scheme encouraging early preparation for that law. Signing it does not satisfy any Article of the Act on its own.

Does the EU AI Office enforce the AI Pact?

No. The AI Office enforces binding general-purpose AI model rules under the Act. The Pact is a separate, non-binding Commission initiative open to any organization and is not policed by the Office.

Is it worth signing the AI Pact for a company with fewer than 50 employees?

Often yes, mainly as a low-cost forcing function. Smaller Mittelstand companies can use the three pledges as a free checklist to start an AI inventory and basic governance charter before the Act’s high-risk obligations become mandatory.

What does signing the AI Pact cost, and do we need our own IT resources?

There is no fee to join. Internal effort depends on how many AI systems you run; most signatories use existing staff to draft the governance strategy, inventory tools, and run basic literacy sessions.

How does the AI Pact connect to how Superkind’s AI agents are deployed?

Superkind’s AI agents operate under customer-controlled approval and data governance from the outset, so the agent inventory and oversight habits a Pact pledge encourages map directly onto systems a company already uses day to day.

Building better software Contact us together