TELUS tests 34 AI models against 620,000 adversarial attacks

TELUS Digital's new benchmark is a useful safety story because it looks at models as production assistants rather than isolated lab objects. The company says Fuel iX Fortify Applied AI Research tested 34 models from 10 providers across more than 620,000 adversarial attacks and 15 vulnerability categories. Attack success rates ranged from 1.3% to 92.9%, and the strongest predictors of safety were reasoning capability and model size. Reasoning models were vulnerable to 19.9% of attacks versus 55.1% for non-reasoning models, while small models of 10B parameters or fewer failed 86% of the time. TELUS also warns enterprises spend roughly $1 on AI trust and security for every $735 invested in AI capabilities. Watch whether buyers start demanding continuous red-team evidence before deploying agents in finance, healthcare, and customer operations.

Key details: TELUS Digital, GenAI Safety Model Benchmark, May 26, 2026, 34 AI models, 10 providers, 620,000+ adversarial attacks, 15 vulnerability categories, 1.3%-92.9% attack success rates.

Continue swiping for more AI Brief stories.