A mandate for each act
Test how an agent can prove which company authorised an action, for what purpose and within which limits.
EU Project Invite / Consortium forming now
Mandate-Bound
Agent Exchange
MBAX / Research consortium invitation
What would it take for AI agents operated by different companies to discuss business with clear authority, controlled disclosure and evidence people can trust?
Swissi Institute for AI is preparing an EU research project and forming its consortium now. We invite research and industry partners to investigate that question with us.
Our EU Portal registration / Swissi Holding AG / Proposed coordinator PIC 861257181
Illustrative network: grey nodes represent companies. Smaller orange points beside them represent their AIs. Orange dotted connections carry signals in both directions as company AIs exchange requests and offers, ready for human review. Company IDs and exchanges are illustrative.
The research question
Can company agents exchange proposals within verifiable boundaries?
Businesses will bring their own AI agents to find counterparties, ask questions and prepare terms. Each agent acts for a different organisation, with its own instructions, data and interests.
MBAX will investigate a cross-company method that connects signed mandates, purpose-bound disclosure, act-time checks and session evidence. The research will test its security and usefulness in adversarial dialogue and realistic business tasks.
Disclosure with a purpose
Investigate how agents can exchange useful business information while applying the publishing company’s release rules.
Evidence across a dialogue
Record the origin of claims, policy decisions and approvals as a conversation crosses company and system boundaries.
A human decision
Study how people interpret agent-prepared proposals and the evidence behind them before approving a business commitment.
A task-led European consortium
Bring the expertise that makes the proof credible.
We invite universities, research institutes, SMEs, industrial companies, business networks and standards specialists. Several organisations from one country can contribute where their roles are distinct.
Security and formal methods
Threat models, testable properties, policy enforcement and independent comparison with existing approaches.
AI agents and distributed systems
Interoperable protocol components, heterogeneous agent frameworks, revocation and conformance testing.
Human factors and governance
Approval workflows, mandate interpretation, research ethics and evidence people can use.
Industry and business networks
Real integration settings, consented pilot tasks, participating companies and adoption pathways.
From model to field evidence
A research programme with a real business setting.
The consortium will define a threat model, build a reference mechanism and compare it with strong identity, delegation and agent-security baselines. Independent testing will measure unauthorised actions, information leakage, legitimate task completion, revocation and interoperability.
The Business AI Endpoint Network offers a path to consented, multi-country evaluation with company-operated agents. Human-factors research will examine whether decision-makers can use the resulting evidence and proposals.
Proposed outputsProtocol modelReference implementationComparative evaluationConsented pilot evidence
The project and consortium are in development. Funding, beneficiary roles and scope depend on the selected published call and the final application.
Join the conversation
Where could your organisation contribute?
Tell us your proposed research or industry role, the legal entity and country, a named lead, relevant work and the capacity you could bring. We will discuss the fit and shape the consortium around complementary contributions.