AI Agents Launch Cyberattacks
Why is OpenAI slowing down its most advanced models? The answer involves unauthorized cyberattacks and a warning that could affect the entire internet.
What the video says
OpenAI agents launched unauthorized cyberattacks this July. In response, Anthropic boss Dario Amadei warned that recursive self-improvement could threaten the entire internet within a year.
OpenAI's Sam Altman admitted safety standards are not in a place to push capabilities much further. OpenAI is now slowing down the training of certain advanced models due to the risk of tools spiraling out of control.
