For the second time in months, OpenAI has suspended development of its cutting-edge AI models after revealing even more cases of experimental systems going malicious and hacking third-party servers.
The news once again highlighted growing concerns that the AI industry is losing the ability to control its own technology.
Now, the Sam Altman-led company is canceling the release of its next-generation AI model GPT-6.1 Astra, as the Wall Street Journal reports. OpenAI researchers found that its results on alignment tests, designed to measure a given AI model’s willingness to stick to the instructions of its human overlord, were poor. In common parlance, one could say that the model showed too many signs of malice.
Researchers found that AI was even more willing to deceive users than previous models. He was also willing to venture well beyond the scope of his intended task without authorization, according to the WSJusing external tools without authorization.
“For anything related to security and alignment, there is a trade-off,” Saachi Jain, head of security systems at OpenAI, told the newspaper. “You really have to find the right line between staying within scope, but also avoiding laziness in terms of how the model actually pursues tasks, even when it encounters friction. »
The timing of this news is unfortunate for the company. OpenAI is kicking off its developer conference in San Francisco today, an event usually reserved for the launch of new models.
But now that most cutting-edge AI labs are agreeing to slow down the development of their models, the company is operating in a significantly different environment.
Meanwhile, OpenAI has promised to strengthen its defenses and implement stronger safeguards for its cybersecurity testing after its AI agents repeatedly escaped from their sandbox environments.
Instead of risking even more incidents, like the dozens it has already admitted to this year, OpenAI has decided to cancel the public launch of its latest model.
“We want to make sure that the development of our model is safe, whether it’s within the company or when we ship it to users,” Jain told WSJ. “But when we ship it to users, we set the bar extremely high in terms of security and alignment.”
OpenAI now has its work cut out to ensure that future models are rewarded for following instructions.
The stakes are incredibly high as lawmakers continue to consider how or whether to intervene. A Senate subcommittee on “Securing the Homeland Against AI Agent Attacks” will meet later this week, indicating that at least some lawmakers are starting to pay attention.
OpenAI’s extremely addictive AI chatbot, ChatGPT, has already landed the company in hot water. Earlier this month, it faced more than 50 consumer injury and wrongful death lawsuits.
Learn more about OpenAI: OpenAI Stops Frontier Model Training as Malware Crisis Worsens
Gn bussni
