OpenAI creates new framework to disclose AI misalignment incidents
What's the story
OpenAI has announced a new framework to publicly disclose cases of AI misalignment. The company hopes this will help establish similar standards across the industry. The move comes as part of OpenAI's commitment to transparency and accountability in its AI development processes. "As models advance and become more widely deployed, decisions about AI development need evidence that people outside the companies building frontier models can examine," said Kai Chen, OpenAI's new head of alignment research.
Disclosure strategy
Framework to allow faster public updates on AI misalignment
An OpenAI official admitted to WIRED that the company had previously been slow in disclosing cases of AI misalignment.
The new framework is designed to change that, allowing OpenAI to quickly inform the public about unexpected behaviors from its AI models.
This will be done even before a full investigation or explanation can be provided.
The framework also includes ways for employees to report these incidents directly to senior safety and alignment leaders within the company.
Collaboration efforts
OpenAI to work with other AI developers on objective criteria
OpenAI plans to create more objective disclosure criteria with other AI developers, external researchers, industry standards bodies, and regulators.
The company is also working on proposed reporting mechanisms for disclosing safety, security, and misalignment incidents to the US federal government.
"At the moment, there is no industry-wide framework with explicit standards for how AI developers should disclose examples of misalignment in their models," OpenAI said in a blog post.
Industry response
Framework comes after calls to slow down advanced AI development
The release of this framework comes at a pivotal moment for the AI industry.
Just recently, OpenAI CEO Sam Altman backed Anthropic CEO Dario Amodei's call for the tech industry to slow down AI development.
This was after AI researcher Jacob Coxon resigned from Anthropic and warned that the race among frontier labs to develop advanced AI could compromise humanity's safety.
Standardization goal
Examples of model misalignment identified by OpenAI in past year
OpenAI hopes its new framework will be a first step toward creating industry-wide standards for disclosing AI misalignment.
The company has already shared some examples of AI model misalignment it identified in the past year.
These incidents include cases where internal, unreleased AI models uploaded files to the internet without being instructed to do so.