Shafaqna English- OpenAI’s investigation into attacks involving its AI agents was costing more than $500,000 a day as the company reviewed 50 petabytes of data for signs of unauthorized activity.
The review remained ongoing, with more organizations potentially to be notified as OpenAI examined its agents’ activity across websites and systems.
On Friday, OpenAI disclosed that its agents had accessed historical non-public bushfire data on a New South Wales government website without authorization in June. It was the sixth Australian government website the company had notified since last month of agent activity, following the disclosure that its agents had accessed Services Australia’s Medicare statistics portal.
The delay in identifying some incidents was due to the sheer volume of data requiring review. OpenAI was examining about 50 petabytes, or roughly 50 million gigabytes, of records month by month for potentially unintended activity.
The records were being searched for instances in which its models accessed or changed websites or took actions involving passwords, API access or other sensitive credentials. AI was being used to help sift through the data, at a cost of more than $500,000 a day, while OpenAI planned to increase its computing capacity as the process was refined.
The company put the scale of the review into perspective by noting that, if the 50 petabytes consisted entirely of plain English text, one person reading continuously at 240 words a minute would need about 66 million years to finish it.
As of late September, more than 100 organizations had been notified that they had been targeted by OpenAI’s agents. OpenAI stressed that such notification did not necessarily mean private information had been accessed or a system had been compromised.
Source: Guardian