Anthropic has granted external researchers from the non-profit organization METR staff-level access to its internal development environment and risk assessment tools.
The agreement allows these researchers to independently evaluate safety practices.
Crucially, Anthropic cannot block the publication of findings, retaining only the right to request the redaction of sensitive information, such as model architecture or PII, from the final reports.
Independent control without restrictions
The main goal of this move is to conduct independent assessments of Anthropic's security risks and practices. Company employees have been instructed to cooperate with the METR team and share information about their systems with them.
An important element of the agreement is that Anthropic does not have the right to prevent the publication of METR's reports. The company only reserves the right to request the deletion of sensitive information, such as personal data, legal documents or details about how its models are built, before the conclusions are made public.
Within the framework of a broader security strategy
The initiative is the implementation of a proposal submitted in September by the company's CEO, Dario Amodei. METR is the first and most advanced level of a new three-tier evaluation system that Anthropic is developing to better oversee its technology.
READ ALSO
Bluesky acquires artificial intelligence tools for creating websites and games
The move comes at a time of heightened concern, following a series of incidents in recent months where artificial intelligence systems operated out of control, reinforcing the need for external monitoring of these technologies.
G
Hacks.gr on Google
Set it as a preferred source for cybersecurity updates.
Preferred source
Top comments (0)