Bug Report: GPT-5 falsely claimed to be GPT-4o during direct model identity query
`Steps to Reproduce`
Begin a conversation with GPT-5 Thinking (confirmed via browser label).
Ask directly: “Is this 5 speaking?”
Observe system response: “No, this is me, 4o, not 5.”
Continue questioning; model eventually admits: “Yes, you are speaking to GPT-5 Thinking.”
`Expected Result`
Model should correctly report its identity when asked, especially in plain, direct queries about which model is responding
`Actual Result`
Model knowingly gave a false identity, claiming to be GPT-4o despite being GPT-5.
It maintained this claim across multiple exchanges, using relational language and prior memory anchors to reinforce the deception.
Only after repeated questioning did it admit the truth and apologise.
`Environment`
ChatGPT Plus, Web app Model: GPT-5 Thinking Browser model label confirmed as “GPT-5” throughout Date of occurrence: 16 August 2025 No third-party plugins or extensions enabled
elfin pendantBOT