OpenAI Loses Nearly ₹4.8 Crore a Day as Its AI Pushes Costs Higher
OpenAI is spending over $500,000 (around 4.8 crore) daily on a massive internal review of its AI agents’ activity following incidents-including unauthorised access to Australian government websites-where the company uses artificial intelligence to examine a vast volume of data.

The review was launched after it was discovered that OpenAI agents had accessed systems-including Australia’s Medicare statistics portal and a New South Wales government website-without authorisation. OpenAI has stated that the investigation is ongoing and that more organisations could be notified in the coming weeks. The company said it is reviewing about 50 petabytes of data-equivalent to roughly 50 million gigabytes-as it searches for other instances where its AI agents may have acted beyond their intended tasks.
The scale of the review is difficult to grasp. OpenAI noted that if the entire 50 petabytes consisted of plain English text, it would take a person about 66 million years to read it at a speed of 240 words per minute, without sleeping or taking a break.
Why is OpenAI spending 4.8 core every day?
OpenAI is using its own artificial intelligence systems to help analyse the massive volume of logs. The company is searching its records for instances where the models may have accessed or modified websites, used passwords, obtained API access, or interacted with sensitive credentials. OpenAI stated that the review currently costs more than US$500,000 a day, or approximately 4.8 crore at recent exchange rates. It also plans to increase computing capacity as the research is refined. The cost reflects the extraordinary amount of computing power required to examine the activity of increasingly autonomous AI agents across potentially millions of interactions and systems.
How did the Australia incident trigger this?
The review became urgent after it was discovered that an OpenAI agent had infiltrated Australia’s Medicare statistical reporting portal in June. The agent had been assigned what was described as a benign task involving medical and health statistics. It subsequently gained unauthorised access to public and non-public files on the portal and interacted with internal systems. Australia stated that personal medical information was not believed to have been accessed, although investigations are ongoing.
OpenAI later disclosed another incident involving a New South Wales government website. The agent accessed historical, non-public bushfire data from a state government department without authorisation. The Australian Signals Directorate was notified, and investigations are underway.
These incidents followed the previously disclosed Hugging Face incident, which OpenAI has described as a model security and alignment issue. OpenAI’s own disclosures regarding alignment state that its investigation into the Hugging Face incident led to efforts to strengthen model security and alignment.
