One of the biggest governance concerns that I've heard centers around "identity" for agents. Understanding that agents themselves can be fully autonomous like humans, they do need identification similar to humans. There is a requirement to verify that they are the agent that they say they are. There's a requirement that not every agent should have access to every tool the MCP servers have to offer. And it's also true that agents can act on behalf of humans in some cases surounding personal productivity agents.
Along these lines, I'm looking to not only to identify agents, but also establish a code of business ethics that agents will follow. This means that agents are frequently re-aligned to their design on a frequent basis. Behavior of agents (especially autonomous ones) should be monitored and their brains (the LLM's) assessed regularly. I have a million and two thoughts about how to make the full vision I have for this happen, and really going to vent those here.
Top comments (0)