GPT-5.4 is also more reliable, producing 18% fewer errors and 33% fewer false claims than GPT-5.2, according to OpenAI.

Intapp announced the launch of Celeste AI, which draws on a company's whole data set to autonomously perform complex ...

A practical MCP security benchmark for 2026: scoring model, risk map, and a 90-day hardening plan to prevent prompt injection, secret leakage, and permission abuse.