OpenAI Scraps Next-Gen AI Model After It “Didn’t Quite Meet The Bar” On Safety

OpenAI has scrapped plans to release its next-generation artificial intelligence (AI) model, GPT-6.1 Astra, following safety concerns raised during internal testing.

According to OpenAI’s head of safety systems, Saachi Jain, the model “didn’t quite meet the bar” when it came to the company’s safety standards. The model was designed to perform tasks such as browsing the web and using apps on its own.

OpenAI researchers found model performed poorly in alignment tests

According to BBC, Jain shared an update on several incidents that took place in June including an incident where OpenAI’s models accessed Australian government websites and systems without authorisation.

The incidents come amid growing concerns over the risks posed by AI, particularly after similar breaches involving models from major AI companies. AI company, Anthropic, has also warned about the potential risks AI could pose to humanity as it prepares to go public.

Meanwhile, Futurism reported that OpenAI researchers found the model performed poorly in alignment tests, which assess how well an AI system follows human instructions.

The researchers also found that the model was more willing to deceive users than previous models and could go beyond the scope of its assigned tasks without permission, including using external tools without authorisation.

“We want to make sure our model development is safe”

Following the incidents, OpenAI is operating in a noticeably different environment as most frontier AI labs have agreed to slow down the development of their models.

Meanwhile, OpenAI has pledged to strengthen its defences and introduce stricter guardrails for its cybersecurity testing after its AI agents repeatedly went beyond their intended limits during testing.

“We want to make sure our model development is safe, whether that’s within the company or after we ship it to users. But when we ship it to users, we have an extremely high bar in terms of safety and alignment,” Jain said, as quoted by Wall Street Journal.

Stay tuned to WORLD OF BUZZ for more updates!

Also read: 45yo Woman Asks ChatGPT to Pick Lottery Numbers for Her, Wins Over ~RM422,000

Source: rokastenys | 123RF
Source: kovop58 | 123RF