Research

Glossary

Plain-language definitions of the terms used across this site, so that the words in evaluations, guides, and comparisons mean one thing.

Last updated 2026-10-10

Agentic workflow: A process in which an AI system plans and carries out several steps toward a goal, calling tools and acting on a user's behalf with some autonomy, instead of answering a single prompt. The degree of autonomy, and the controls placed on it, vary by product.
Authorization boundary: The defined scope of systems, services, and deployment to which a security authorization applies. A status covers only what falls inside that boundary, so it should never be assumed to extend to a vendor's other products.
Benchmark contamination: When material used to test an AI system was also present in its training data, so a high score may reflect memorization rather than capability. It is one reason public benchmark results can overstate performance on genuinely new tasks.
Hosted deployment: A deployment where the vendor runs and maintains the software on its own infrastructure, so the customer sends data to the vendor's service and relies on the vendor for operations, updates, and security.
Independently measured result: A result produced under a documented, repeatable protocol by someone independent of the vendor, with the conditions, product version, and date recorded. It is the strongest evidence class used on this site.
Insufficient evidence: The status shown when a dimension lacks independent evidence that meets the methodology's minimum rules. No score is shown, and the gap is never filled with an average or neutral value.
Self-managed deployment: A deployment where the customer runs the software on infrastructure it controls, such as its own servers or cloud account, and takes on operations, updates, and security responsibilities.
Superintelligence: A hypothetical level of machine intelligence that greatly exceeds human performance across most areas. Definitions differ among researchers and institutions, and the term does not describe any product sold today.
Vendor-reported claim: A statement made by the company that sells a product about its own product. It is a lead to check, not a finding, and this site always labels it as vendor-reported.