Red-Team Testing for Public-Facing AI Agents
Public-facing AI agents need to be tested against more than expected questions and cooperative users.
Happy-path testing is only the beginning
A lot of agent testing starts with straightforward checks:
Can the agent answer the question?
Can it use the correct knowledge base?
Can it complete the requested task?
These checks are useful, but they do not reveal how the agent will behave under the unpredictable conditions created by real users.
What else should teams test?
Before an agent is deployed publicly, teams should also examine how it responds to:
attempts to bypass its policies;
situations in which escalation is required;
messy, ambiguous or incomplete inputs;
long-running, multi-turn interactions;
changes to prompts, models, tools or workflows.
These tests help uncover cases in which an agent appears reliable under normal conditions but behaves differently when it is pressured or confused.
We have been expanding Spec27’s red-team testing support to make these checks easier to run, review and repeat.
Moving beyond early access
At the time of this update, Spec27 was also moving beyond early access with Free, Plus and Pro usage levels.
The Free level remained available, while teams could request access to the additional levels and understand where their usage would sit before paid access began. Existing accounts received Plus-level credits and usage permissions.
This update was originally shared on 1 September 2026, when the Spec27 team was attending AI Enterprise Conference NYC and preparing for AI in Financial Services London.
For more practical guidance, read Red-Teaming Public-Facing Agents.
