NEW YORK, NY — A senior consumer staples analyst at a midtown investment bank fed ChatGPT-5 his complete research methodology Wednesday morning, requested a fresh Q3 sector outlook, and received a 14-page report identical to one he filed in October 2019, down to the same Procter & Gamble price target and an uncorrected typo on page nine reading ‘strong long-term momenntum.’
The analyst, who spent the next forty minutes re-prompting the model while a tuna salad sandwich slowly browned on top of his desk calendar, escalated the matter to OpenAI’s enterprise support line. A representative responded within the hour that the model was ‘behaving as designed’ and offered a 10% discount on additional API credits.
‘ChatGPT-5 has been trained on the full corpus of sell-side equity research, and what it learned is that nobody changes their mind,’ said Hollis Kerrigan, a partner at Beacon Strategy who advises Tier-1 banks on AI deployment. ‘The model is not regurgitating his note. It independently arrived at the exact same note, holding the same view, with the same rationale. Which, frankly, is the only honest review of sell-side research ever produced.’
By Wednesday afternoon, four other analysts on the same floor had run comparable tests, with comparable results. A healthcare analyst received a 2017 managed-care outlook in which the model had replaced every instance of ‘Obamacare’ with the word ‘Obamacare.’ A semiconductor analyst was issued a Buy rating on Intel dated March 2014, accompanied by a one-line note from the model reading, ‘Just keep filing this one, it works.’
An energy analyst, attempting to stress-test the system, asked ChatGPT-5 to generate a maximally contrarian outlook on the integrated majors. The model returned his existing Buy rating with the price target raised by one dollar and a new sentence inserted in the executive summary: ‘We continue to see asymmetric upside.’
Bank leadership, briefed on the findings over a working lunch of cold chicken Caesar wraps, announced Wednesday evening that ChatGPT-5 would be deployed across the equity research division by Q4, citing ‘demonstrated alignment with the firm’s existing analytical framework’ and ‘an opportunity to right-size headcount by approximately the entire equity research division.’ Junior analysts learned the news in a 4:45 p.m. meeting catered with leftover pastries from the morning’s all-hands, where the head of research praised the model’s ‘institutional memory’ and its ‘refusal to be rattled by short-term noise, or, you know, evidence.’
OpenAI declined to comment on the specific incident but pointed reporters to a recent blog post in which the company described GPT-5’s ‘enhanced consistency over long contexts’ as one of the model’s flagship improvements. A spokesperson added that the model had been specifically fine-tuned ‘to reflect the conventions of professional financial communication,’ a sentence that Kerrigan, reached for follow-up, said was ‘the most damning thing OpenAI has ever published, if you read it correctly.’
The analyst at the center of the incident has reportedly accepted the situation. He spent Thursday morning copying the regenerated note into the bank’s research template, sent it to compliance for the same review it received in 2019, and added the document to a distribution list of 1,400 institutional clients, none of whom are expected to notice. According to people familiar with the file, the model’s sole original contribution to the report was a footnote on the final page that read, in plain unbolded text, ‘I notice you have been writing this same note for six years. Have you considered speaking with someone.’