The agent is verbalizing internal conditional logic and prompt instructions during live calls, exposing system-level decision-making processes to end users. Specifically, the agent is speaking aloud bracketed conditional statements such as "[If they don't remember]", "[If they're confused]", and "[If they want to be removed]" that should remain internal to its reasoning process and announcing the current time which should be stored as a variable only. Please help us fix this immediately https://cdn.discordapp.com/attachments/1472769523479482450/1472769524032995350/image.png?ex=6993c701&is=69927581&hm=54ae536a957db8adef674e2ceb3f2dc1f9fccf3e5a33a424a36dccff11df3030& https://cdn.discordapp.com/attachments/1472769523479482450/1472769524485849232/image.png?ex=6993c701&is=69927581&hm=5da8dfda4ddcf49146723fda98dab7c41ade79a46d12ec161f7a27a333c7feb8&

@teppie hey, what model is this?

Initially we were using GPT 4.1 but then it upgraded to GPT 5 https://cdn.discordapp.com/attachments/1472769523479482450/1472900346408996894/image.png?ex=699440d7&is=6992ef57&hm=5d552dc0bdb22838628ecf715a6eed3a123d444107cfbf87b84390dba8ef18e5& https://cdn.discordapp.com/attachments/1472769523479482450/1472900346773766227/image.png?ex=699440d7&is=6992ef57&hm=830c8956fd681e5a6491c8da9f150afbfa87e6590be8297bae6c391391794cac&