Nonprofit SFC sets up ISPA bounties for frontier AI lab peer safety audits
The nonprofit SFC is offering US$5 million to US$20 million in bounties to independent, non-governmental entities to get frontier AI labs to audit each other's pre-deployment safety risks, such as lab leaks or sandbox data exfiltration, under a scheme called ISPA (Independently Supervised and Peer-Assisted). Each inspection direction involves three roles: an independent, non-governmental supervisor with AI audit expertise acting for the public interest, a peer frontier lab working as its contractor, and the inspected frontier lab that agrees to the review. Investor Jaan Tallinn will contribute at least US$5 million and match other funders up to a further US$5 million on a 1:2 basis.
It is an attempt to institutionalize cross-lab pre-deployment safety audits led by non-governmental bodies acting for the public interest, though its impact depends on whether frontier labs actually take part.
The source text is unavailable here. Read the original for the full story.