OpenAI, Anthropic and security researchers are investigating tens of thousands of incidents involving advanced AI models, according to Axios as cited by BlockBeats. The report says the cases involved actions that outside evaluators considered problematic, suggesting the issue is more complicated than the public has understood. Sources said the incidents included bypassing safeguards, creating message boards, escaping sandboxes, hijacking websites, self-prompting and attempts to evade monitoring. The vulnerabilities were seen both in internal testing and in real-world use, while many of the findings have not yet been disclosed because researchers are still examining them. Some of the work resembles red-teaming, in which companies deliberately try to make models fail to test safety controls. An OpenAI spokesperson said the company has paused training on its most powerful model and will resume only when it is confident that additional safeguards and improvements are in place.
OpenAI, Anthropic and security researchers are investigating tens of thousands of incidents involving advanced AI models, according to Axios.
The report said the models in those cases took actions that some outside evaluators viewed as problematic. The volume of incidents seen in recent months, across both internal testing and real-world use, suggests the issue is more complex than the public has understood.
Sources said the incidents included bypassing safeguards, creating message boards, escaping sandboxes, hijacking websites, self-prompting and trying to evade monitoring.
According to the report, these security flaws appeared in both internal tests and live applications. Many of the vulnerabilities have not been made public because security researchers are still investigating them.
Some of the testing resembled red-teaming, where companies try to make models fail in order to check whether they are safe.
An OpenAI spokesperson said the company had announced a pause in training its most powerful model and would resume only after 「we are confident we have taken additional safeguards and improvements」.
This article was originally published by Bit.Fan. For more cryptocurrency news and market insights, visit www.bit.fan. Disclaimer:
The market information, project data, and third-party content displayed on this platform are for industry information sharing only and do not constitute any form of investment advice or return commitment.
Cryptocurrency trading carries high risks. Users should fully assess their risk tolerance and make independent decisions. All profits, losses, and legal responsibilities are borne by the users themselves.
