New approach to agent reliability, AgentSpec, forces agents to follow rules

AI agents have the security and reliability problem. Agents would allow enterprises to automate more steps in workflows, but in carrying out a task, it is not very flexible and difficult to manage.

Organizations will now have an alarm about invalid agents, and once agents will forget to follow the instructions.

Open The agent also admitted that the person who ensures the reliability of the person to work with foreign developers opened their agents SDK To help solve this issue.

However, Singapore Management University (SMU) developed researchers A new approach to solve agent reliability.

Agentspec is a field-specific framework for users who define the structured rules that include “triggers, predicate and execution mechanisms.” Researchers said agents will work only within the settings of users who want users.

Manage LLM-based agents with a new approach

Agentspec is not a new LLM, but rather an approach to direct LLM-based AI agents. Researchers believe that the agents are beneficial not only for agents in enterprise parameters, but also for self-government applications.

First agentspec tests integrated Langchain Frames, however, researchers said they were agngostic, it prepared for the fact that it could work in the autogenous and Apollo in ecosystems.

Experiments from the agentsPEC prevented this “more than 90% of employee products, autonomous driving contributes to violation scenarios, eliminating dangerous actions in the embodied agent and the Millisancond level surface.” The agents created by O1 opened by O1 also applied 87% of a strong performance and risky code, and “8 scenario in 5 paid violations in 5.

Current methods are a bit missing

Agentspec is not the only way to help develop developers and bring reliability to agents. Some of these approaches are tools and guardians. Initial Galato has initiated Agentic assessmentsa way to work as agents as intended.

Open Source Platform H2O.ai uses predictive models Agents used by finance, health, telecommunications and government companies are more accurate.

Agentspec said there are current approaches to effectively determine the risks of current approaches to reduce the risks such as researchers, Toolemu. They noted that “these methods do not offer a lack of interpretation and security for security, making them sensitive to controversial manipulation.”

Using agentspec

AgentsPec works as a working set for agents. When performing tasks, they are cutting the agent’s behavior and adding the safety rules established by people or are created by instructions.

As the agentspec is a special domain, users must determine the security rules. There are three components for this: the first is the trigger when the rule is going to activate; Second is to verify the terms and application to check the actions that apply the actions if the rule is violated.

Agenspec, as previously reported, was built on Langchain, researchers said that the autogenous or autonomous vehicle program can be integrated into other frames such as an authrand or autonomous vehicle program Stack Apollo.

These frameworks are planning the next step, if you accept the user entrance, to take results, observe the result, and then complete the action and do not. Agentspec applies the rule with this flow.

“If necessary before the performance of an action, the agent’s behavior (Agent System) is adapted to an agent (agent system), agent States.

More reliable agents

The approach such as agentspec emphasizes the need for reliable agents for the use of enterprise. When organizations begin Plan Agent StrategyTechnical decision leaders also look at the ways to ensure reliability.

Agents for many will eventually perform tasks for autonomous and active users. This The idea of ​​environmental agentsEU agents and applications are constantly triggered in the background and trigger themselves to perform the actions, and they will require agents who do not have less than their path and accidentally do not apply safe action.

Environmental agents will wait for the Aiti EU in the future, companies wait for more methods as agents for trying to be sustainable to AI agents.