As AI agents start acting on each other’s behalf across institutions, knowing who you’re really talking to becomes the question that Sparkle is answering with its new STLS-AI service
Imagine a researcher who wants to analyse data held by a partner university: their AI agent contacts an agent at that university to request access, then works with an agent at a research computing centre to arrange the resources for the analysis. What could have been a string of emails and approvals becomes a coordinated workflow, leaving the researcher more time for the scientific question itself.
Let’s focus on the moment the first request arrives: the university is being asked to release research data. Who is the agent asking for it, and on whose authority? The requester, in turn, needs to know it is dealing with the genuine partner. When the work reaches the computing centre, that institution must establish whether the agent requesting its resources is entitled to use them.
This is the trust problem behind agentic AI. Agents are software systems that break a task into steps, use tools, assess results and decide what to do next, including asking other agents to do part of the work. Across a research consortium, many such exchanges happen through shared platforms, and some agents are ephemeral: created for one task, retired once it’s done. As agents multiply and tasks change, so do the relationships and permissions that govern them. A single mistake could expose unpublished data or let resources be used without permission. The more work agents carry forward on their own, the more precisely institutions need to control the relationships behind that work.
Sparkle is developing STLS-AI (Symmetric TLS for Agentic AI) to answer exactly this. It is a patent-pending solution, built on Sparkle’s experience in secure communications, that lets agents recognise one another and establish protected, independently manageable relationships that institutions can govern directly.
STLS-AI derives each agent’s cryptographic identity from symmetric-key cryptography, shared secret keys that let an agent verify it is talking to the expected counterpart. That recognition is the basis for a protected relationship that institutions can establish, renew or withdraw as a collaboration evolves, without disrupting others running on the same platform. That lifecycle control matters most for agents that exist only for a single task, and STLS-AI is built to manage identity and trust at the scale that agentic collaboration demands.
Now let’s go back to our researcher’s case: the participating agents verify each other’s cryptographic identity and protect their exchange, while the university’s own identity and access systems decide which institution the requesting agent represents and whether it may receive the dataset. If the collaboration ends, the university can withdraw trust from that one relationship while everything else on the platform continues.
Agents could help carry scientific work further, across more systems and institutions. The authority to decide who takes part must stay with the people and institutions behind that work.
STLS-AI is also designed to support auditability, tracing agents’ identities, protected sessions and changes in trust so institutions can reconstruct what happened without inspecting the research content itself. For institutions that need to prove compliance years after a project ends, this matters as much as the encryption itself. And because sensitive data can outlive any single collaboration by years, STLS-AI combines symmetric encryption with quantum-safe key establishment, building resistance to future quantum threats directly into the design.
Sparkle is already applying with success this protected framework in its own agentic internal infrastructure, and demonstrations are available to show how the STLS-AI approach could support research collaborations with their different participants, platforms and institutional rules.
To learn more, join the forthcoming STLS-AI webinar on 23rd November by Sparkle – in partnership with GÉANT – dedicated to the research and education community, introducing the approach and walking through a representative agent-to-agent collaboration.
Register Here
