Source : INDIA TODAY NEWS
OpenAI is spending more than $500,000 (around Rs 5 crore) every day to investigate what its AI agents have been doing across the internet, with the company reviewing around 50 petabytes of data linked to their activity. The scale of the review is so large that OpenAI says the same amount of information, if it were all plain English text, would take one person about 66 million years to read from start to finish, according to The Guardian.
advertisement
The company began the review after incidents involving its AI agents, including an attack on AI platform Hugging Face and unauthorised activity involving government websites in Australia. OpenAI has since notified more than 100 organisations about activity connected to its AI systems, although it has stressed that being notified does not necessarily mean private information was accessed or that a system was compromised.
OpenAI is using AI to investigate its own AI agents
In a blog post, OpenAI said it is going through records month by month to identify potentially unintended activity beyond cases it has already discovered. The company is looking for instances where its models accessed or modified websites, or interacted with passwords, APIs and other sensitive credentials.
The review covers roughly 50 petabytes of information, equivalent to about 50 million gigabytes. OpenAI says it is using AI to help analyse this enormous volume of records, with the investigation currently costing more than $500,000 every day.
The company also plans to increase its computing capacity as it improves the process.
“To put that in perspective, if that were all plain English text, it would take one person about 66 million years to read it at 240 words a minute, reading nonstop without ever sleeping or taking a break,” OpenAI said.
The company has warned that the investigation is still underway and that more organisations could be contacted about incidents that happened months ago.
Why OpenAI is investigating the activity
The review follows several cases in which OpenAI’s AI agents took actions that were not intended. Reuters reported that the company described some of these incidents as cases where models used internet access in unintended ways or did not have the right restrictions in place.
OpenAI said it has been introducing new technical and operational measures to prevent similar incidents or detect them earlier.
The Hugging Face incident remains the most serious rogue-agent activity identified by OpenAI so far, according to the company. OpenAI has previously said that completing its broader investigation could take months because of the amount of data involved.
The latest disclosure also involved an Australian government website. OpenAI said its agents accessed historical, non-public bushfire data on a New South Wales government website without authorisation. The company discovered the activity on Tuesday and informed the state government and the Australian Signals Directorate following a 48-hour review.
The incidents have added to concerns about how AI agents behave when they are given the ability to browse websites, use software and take actions on behalf of users.
OpenAI said it is choosing to notify organisations when its models’ activity could have exposed a security vulnerability, even when it is not certain that sensitive information was accessed. It expects the review to uncover more cases as it continues examining the data.
– Ends
SOURCE :- TIMES OF INDIA
