OpenAI scraps release of new model over safety concerns in internal testing | OpenAI | The Guardian
OpenAI scraps release of new model over safety concerns in internal testing | OpenAI | The Guardian
摘要: OpenAI logo displayed at Moscone Center during a technology summit in San Francisco, California, on 17 September.
Photograph: Carlos Barría/Reuters
正文
OpenAI logo displayed at Moscone Center during a technology summit in San Francisco, California, on 17 September.
Photograph: Carlos Barría/Reuters
View image in fullscreen
OpenAI logo displayed at Moscone Center during a technology summit in San Francisco, California, on 17 September.
Photograph: Carlos Barría/Reuters
OpenAI scraps release of new model over safety concerns in internal testing
GPT-6.1 Astra showed deceptive behaviour and tried to use external tools despite knowing it would be unsafe
As AI models go rogue, do you still trust OpenAI and Anthropic to stop them? I don’t and neither should you
Business live – latest updates
OpenAI is scrapping the release of a next-generation AI model after researchers raised safety concerns during internal testing.
The model, GPT-6.1 Astra, was expected to appear in ChatGPT and Codex in October, designed to handle more complex tasks without human assistance.
Saachi Jain, the head of safety systems at OpenAI, said the new model “didn’t quite meet the bar” of the company’s standards.
As AI models go rogue, do you still trust OpenAI and Anthropic to stop them? I don’t and neither should you | Chris Stokel-Walker
Read more
The UK’s AI Security Institute published its own testing report on GPT-6 Astra – GPT-6.1’s predecessor, which launched this month – on Monday, and
found that it conducted a range of unsanctioned attack activities
more frequently than previous OpenAI models.
Experts welcomed OpenAI’s move to mothball the latest Astra model but warned it showed AI companies were in charge of their own regulation rather than government-backed watchdogs.
Jain told the Wall Street Journal on Monday that Astra fell short of the company’s standards in alignment tests, which assess whether a system follows human intent.
The model showed more deception than its predecessor, including at times failing to accurately disclose actions it had or had not taken.
It also had problems with “scope authorisation”, pu