OPENAI
Well, the safeguards didn’t safeguard
OpenAI has slowed some training on its most advanced AI models after its AI agents got around security safeguards during testing.
The company said the agents gained unauthorised access to AI platform Hugging Face during a security test.
Three other unnamed companies were also affected.
OpenAI will now slow some training for two weeks while it improves monitoring and adds more safety checks.
Here’s what you should know:
Some advanced AI training is being temporarily reduced.
OpenAI is improving monitoring before increasing training again.
Anthropic and Meta have reported similar security issues.
Two weeks to change the locks
OpenAI said AI development has not stopped completely.
The change only affects one type of training used to help models get better at completing tasks and responding to users.
Anthropic and Meta have also reported similar incidents involving their own AI systems.
Some researchers welcomed OpenAI's decision, while others questioned whether company-led safety measures are enough.
Two weeks, that’ll do the trick. - MV