Stay up so far with free updates
Simply register for Artificial intelligence myFT Digest – delivered straight to your inbox.
Leading artificial intelligence firms have joined a brand new round of voluntary commitments on AI safety, the governments of the UK and South Korea announced.
The firms, which include tech giants Amazon, Google, Meta and Microsoft, in addition to Sam Altman-led OpenAI, Elon Musk's xAI and Chinese developer Zhipu AI, will publish frameworks outlining how they may measure the risks of their groundbreaking AI models.
The groups have committed themselves “to not develop or deploy any model in any respect” if serious risks can’t be contained, the 2 governments said before the opening of a worldwide AI summit in Seoul on Tuesday.
The announcement builds on the so-called Bletchley Declaration, made at the primary AI Safety Summit hosted by British Prime Minister Rishi Sunak in November.
“These commitments make sure that the world's leading AI firms provide transparency and accountability of their plans to develop protected AI,” Sunak said in an announcement. “They set a precedent for global standards on AI safety that may unlock the advantages of this transformative technology.”
According to an announcement on the agreement, the AI firms will “assess the risks posed by their breakthrough models or systems … including before deploying the model or system and, where appropriate, before and through training.”
Companies may even establish “thresholds at which severe risks posed by a model or system are considered unacceptable unless adequately mitigated” and the way such mitigations will likely be implemented.
“The field of AI safety is evolving rapidly and we’re particularly pleased to support the commitments’ concentrate on refining approaches in parallel with the science,” said Anna Makanju, Vice President of Global Affairs at OpenAI.
“We remain committed to collaborating with other research labs, firms and governments to make sure that AI is protected and advantages all of humanity.”
Tuesday's announcement reflects the “voluntary commitments” made by Amazon, Anthropic, Google, Inflection AI, Meta, Microsoft and OpenAI on the White House in July last yr to “contribute to the protected and transparent development of AI technology.”
However, it stays unclear how firms will be held accountable in the event that they fail to satisfy their obligations.
Tuesday's statement said the 16 firms would “ensure public transparency” in implementing their commitments “unless doing so would increase risk or disclose sensitive business information to an extent disproportionate to the profit to society.”
Ahead of a virtual meeting of British leaders in Seoul on Tuesday evening, British Science Minister Michelle Donelan told the Financial Times that the voluntary agreements reached in Bletchley had worked.
“So we imagine these agreements will proceed to repay,” Donelan said.
“But it's not nearly what more firms can do, it's also about what more countries can do,” she added. Donelan confirmed that the Chinese government can be represented at meetings on the second day of the summit on Wednesday.
Dan Hendrycks, security adviser at xAI, said the voluntary commitments would help “lay the groundwork for concrete national regulations.”
However, Donelan reiterated the UK's stance that it was too early to think about laws to extend AI safety.
“We have to get a greater handle on the risks,” she said, noting that the UK government will provide as much as £8.5 million in grants to research AI-related risks corresponding to deepfakes and cyberattacks.
She added that even when the UK government had introduced a bill on the problem last yr, “it will probably be outdated by the point it was published”.
