The Financial Times reports that OpenAI has significantly reduced its safety testing timeline to just a few days, compared to the months typically allocated. This has raised alarms among staff and testers regarding the thoroughness of evaluations that are critical for identifying potential risks, such as security vulnerabilities in new models. Sources attribute this rush to OpenAI's need to stay competitive against emerging open-source models, reflecting a trend toward less rigorous safety practices amidst a growing AI landscape with minimal regulatory oversight.
"We had more thorough safety testing when [the technology] was less important," one person, who is currently testing o3, the full version of o3-mini, told FT.
Sources added that OpenAI's tests are not as thorough as they used to be and lack the necessary time and resources to properly catch and mitigate risks.
Collection
[
|
...
]