Sentient Index Labs Rolls Out Independent AI Behavior Risk Assessment

Sentient Index Labs & Technology (SILT) has launched the Sentience Evaluation Battery (S.E.B.), a new independent assessment tool for artificial intelligence systems. The S.E.B. evaluates AI behavior, focusing on emergent autonomy, deception resistance, and value stability rather than a system's raw capabilities.

The assessment protocol includes 59 adversarial tests across seven behavioral domains. These tests are administered and scored by four independent AI judges operating under blind protocols, with a reported inter-rater agreement of 0.856. Currently, S.E.B. evaluates frontier AI models from companies including OpenAI, Anthropic, and Google.

Results are presented through AI DEFCON ratings, which classify threat levels by comparing a model's capabilities with its behavioral integrity. Additional outputs include an S-Level classification for tracking sentience indicators over time and S.E.B. Projections to monitor behavioral shifts across AI model versions. SILT stated that the data is secured with AES-256-GCM encryption and verification measures.

A key feature of S.E.B. is its declared independence. SILT states it accepts no funding from AI vendors, ensuring its assessments are free from commercial influence. The company also commits to client confidentiality. The S.E.B. data is intended to support compliance documentation for regulations like the EU AI Act and the NIST AI Risk Management Framework.

Evaluations through S.E.B. are available via subscription on the silt-seb.com website.