AI company Anthropic quietly removes 'Declaration to Promote Safe and Trustworthy AI' from its website



Anthropic , an artificial intelligence (AI) startup founded by former members of OpenAI, classifies safety threats associated with improved AI performance using an index called ' AI Safety Level (ASL) ' and predicts that AI that poses a threat to humanity will appear between 2025 and 2026. It places emphasis on the safety and reliability of AI as its corporate philosophy of responsible use of AI. It has been reported that Anthropic has removed from its website a statement that mentioned several voluntary initiatives it created in 2023 in collaboration with the Biden administration.




Anthropic quietly removes Biden-era AI policy commitments from its website | TechCrunch
https://techcrunch.com/2025/03/05/anthropic-quietly-removes-biden-era-ai-policy-commitments-from-its-website/



With the rapid advancement of AI technology, many problems have emerged, such as the damage caused by deep fakes by generative AI and concerns that AI will support weapons development. Therefore, in 2023, the US government requested seven major AI companies, OpenAI, Meta, Microsoft, Google, Amazon, Anthropic, and Inflection, to take 'voluntary steps to mitigate the risks of AI.' The agreement based on this calls for the development of mechanisms for watermarking AI-generated content and for prioritizing research into societal risks such as the promotion of discrimination or prejudice and the violation of privacy.

Major AI development companies such as OpenAI and Google announce that they will work to improve the safety of AI by 'adding watermarks to AI-generated content' - GIGAZINE



As part of its agreement with the government, Anthropic listed its commitments, including information sharing and research, on managing AI risks on a page called ' Transparency Hub: Voluntary Initiatives .'



However, according to Watchtower , an organization that monitors corporate and government AI safety policies, Anthropic quietly removed the section of its 'Voluntary Initiatives for Safe, Secure, and Trustworthy AI' that agreed to the government's request from the 'Transparency Hub: Voluntary Initiatives' without leaving any trace.

Soon after taking office in January 2025, President Trump revoked approximately 80 executive orders from the previous administration, including Executive Order 14110, ' Executive Order on the Safe, Secure, and Trustworthy Development and Use of AI, ' which was issued on October 30, 2023. President Trump also signed an executive order to 'repeal radical and wasteful government diversity, equity, and inclusion (DEI) programs and incentives,' and accordingly deleted pages on which Open AI , Meta , and Google had expressed their commitment to DEI.

President Trump immediately rescinds about 80 executive orders from the previous administration, scrapping AI regulations and LGBT equality initiatives - GIGAZINE



The request made in 2023 is not legally binding, and Anthropic's change to the 'Transparency Hub: Voluntary Initiative' is likely due to a policy change following the inauguration of President Trump, but Watchtower points out that 'the 2023 pledge is not time-limited or dependent on the political party affiliation of the incumbent president.' In addition, while Open AI and Google have explicitly changed their policies, Anthropic has only amended its policy without announcing it.

Watchtower said, 'Anthropic has only removed its 2023 government commitments and has made few other changes to the details of its internal practices, so there is nothing to be concerned about. All Anthropic needs to do now is provide transparency about its choice to abandon its commitments.'

When contacted by web media TechCrunch, Anthropic said in a statement, 'We remain committed to the voluntary AI initiatives established under the Biden Administration. Our progress and specific initiatives continue to be reflected in our content transparency hub. To prevent further confusion, we will add a section that directly identifies where our progress aligns,' explaining that the change to the page does not mean a change in internal policy.

Related Posts:

in Software, Posted by log1e_dh