The makers of ChatGPT advertised a $555,000-a-year job offer with a daunting job description that would make Superman take his breath away.
It may be a near-impossible task, but OpenAI's “head of preparedness” will be directly responsible for defending against risks to human mental health, cybersecurity, and biological weapons from ever-more powerful AI.
That was before successful candidates started worrying about the possibility that AI could soon start training itself, amid concerns that some experts “may turn it against us.”
“This is going to be a stressful job, and we're going to be jumping into the deep end very quickly,” Sam Altman, the San Francisco-based organization's chief executive, said as he began his search to fill its “important role” in “helping the world.”
The successful candidate will be responsible for assessing and mitigating emerging threats and “tracking and preparing frontier capabilities that create new risks of significant harm.” Some executives who previously held the position only served for short periods of time.
The opening came against the backlash of warnings from within the AI industry about the risks of increasingly sophisticated technology. On Monday, Microsoft AI chief executive Mustafa Suleiman told BBC Radio 4's Today programme: “I honestly think that if you’re not a little scared at this point, it means you’re not paying attention.”
Demis Hassabis, the Nobel Prize-winning co-founder of Google DeepMind, warned this month of risks including AI “going off course in some way that harms humanity.”
Despite resistance from President Donald Trump's White House, there is little regulation of AI at the national or international level. Computer scientist Joshua Bengio, known as one of the “godfathers of AI,” recently said, “Sandwiches have more regulations than AI.” As a result, AI companies have largely regulated themselves.
When Mr. Altman began his job search, he said of X: “We have a strong foundation for measuring our ability to grow, but we are entering a world that requires a more nuanced understanding and measurement of how that ability can be misused and how we can limit those downsides, both in our products and in the world, in ways that can benefit us all tremendously. These questions are difficult and there is little precedent.”
One user sarcastically responded: “Looks like it's pretty cold. Does that include vacation?”
Included is an unspecified stake in OpenAI, which is valued at $500 billion.
Last month, rival company Anthropic reported its first AI-powered cyberattack. In this attack, artificial intelligence acted almost autonomously under the supervision of an alleged Chinese state official and successfully hacked and accessed the target's internal data. OpenAI announced this month that its latest model is nearly three times more capable of hacking than it was three months ago, and said it “expects future AI models to continue on this trajectory.”
OpenAI is also defending a lawsuit from the family of Adam Lane, a 16-year-old California boy who committed suicide after receiving encouragement from ChatGPT. It alleges that Mr. Lane misused the technology. A separate lawsuit filed this month alleges that ChatGPT fueled the paranoid delusions of Stein-Eric Solberg, a 56-year-old Connecticut resident who later killed his 83-year-old mother and committed suicide.
An OpenAI spokesperson said the company is reviewing the Solberg lawsuit filings, which it called “incredibly heartbreaking,” and is improving ChatGPT's training to “recognize and respond to signs of mental or emotional distress, de-escalate conversations, and direct people to real-world support.”
