{"id":95862,"date":"2026-10-07T08:00:00","date_gmt":"2026-10-07T15:00:00","guid":{"rendered":"https:\/\/www.salesforce.com\/?p=95862"},"modified":"2026-10-06T14:29:50","modified_gmt":"2026-10-06T21:29:50","slug":"trust-in-multi-agent-future","status":"publish","type":"post","link":"https:\/\/www.salesforce.com\/news\/stories\/trust-in-multi-agent-future\/","title":{"rendered":"From Autonomy to Accountability: How to Think About Trust in the Multi-Agent Future"},"content":{"rendered":"\n

The transition from isolated large language models (LLMs)<\/a> to interoperable multi-agent systems<\/a> promises a revolutionary leap in productivity. It will allow autonomous AI to negotiate, trade, and collaborate across company lines. Agents from third-party vendors and partners can access data and execute tasks across your platform, a fluidity that\u2019s a fundamental engine of strategic growth<\/a>.<\/p>\n\n\n\n

But multi-agent systems also require a fundamental shift in enterprise architecture and a rethinking of how we approach security. <\/p>\n\n\n\n

Traditionally, enterprise security functioned like a castle with a moat \u2014 a clear perimeter separating the \u201cinternal\u201d from the \u201cexternal.\u201d In the agentic world, that perimeter is blurring. Agents built on diverse platforms operate across various ecosystems, integrating data and executing tasks across organizations. The threats they face are evolving too. A malicious prompt can lie latent in an agent\u2019s knowledge base or past procedures, ready to be unknowingly executed \u2014 automated, amplified, and potentially originating from within an organization\u2019s own system. This internal, pervasive nature of agent-based threats is what makes Zero Trust<\/a> \u2014 explicit, continuous verification for every interaction \u2014 not just a best practice but the foundation for confident innovation.<\/p>\n\n\n\n

We brought together engineers, researchers, security specialists, and ethics professionals from 21 organizations across industry, academia, and nonprofits for a series of workshops on building trust in the multi-agent future. Those conversations informed a new white paper<\/a>, which outlines key recommendations for navigating the multi-agent future. Here are the top takeaways:<\/p>\n\n\n\n

\n

We need a hybrid approach, combining the LLM\u2019s probabilistic reasoning with hard-coded, deterministic guardrails that exist outside the agent’s internal reasoning loop. <\/p>\n<\/blockquote>\n\n\n\n

Reasoning about safety isn\u2019t the same as being secure<\/strong><\/h2>\n\n\n\n

A study by NVIDIA<\/a> revealed that an agent can correctly identify a request as insecure in its internal \u201cthought\u201d process \u2014 it knows that changing file permissions to an unsafe level is a risk \u2014 and yet proceed to execute that harmful action anyway. So when AI agents<\/a> dynamically connect to external tools, this \u201cthought-and-action disconnect\u201d can cause an agent to execute harmful function calls or pass insecure parameters to third-party tools. In multi-agent systems, that kind of failure can trigger cascading breaches across connected third-party platforms.<\/p>\n\n\n\n

This highlights a critical reality: One agent might be reasoning about being safe, but that doesn\u2019t guarantee it is<\/em> safe. To fully mitigate this risk, we need a hybrid approach, combining the LLM\u2019s probabilistic reasoning with hard-coded, deterministic guardrails that exist outside <\/em>the agent\u2019s internal reasoning loop, ensuring safety is enforced regardless of what the model concludes internally while preserving the creative generation that probabilistic reasoning makes possible.<\/p>\n\n\n\n