Anthropic Removes Biden AI Policy Commitments - What Happened?

Anthropic Revises AI Safety Commitments

Anthropic has recently removed several previously publicized voluntary commitments from its website. These commitments were initially made in 2023 in collaboration with the Biden administration, focusing on the development of safe and “trustworthy” artificial intelligence.

Details of the Removed Commitments

The withdrawn pledges encompassed sharing information regarding the management of AI risks between industry stakeholders and governmental bodies. Research into potential AI bias and discrimination was also included. According to reports from the AI watchdog group, The Midas Project, these deletions occurred last week from Anthropic’s transparency hub.

Notably, other commitments from the Biden-era, specifically those aimed at mitigating AI-generated image-based sexual abuse, have not been altered.

Lack of Transparency

The change appears to have been implemented without prior notification. A request for comment sent to the company initially received no response.

Original Agreements

In July 2023, Anthropic, alongside other major tech companies like OpenAI, Google, Microsoft, Meta, and Inflection, announced their agreement to follow specific voluntary AI safety commitments. These were proposed by the Biden administration.

These commitments included conducting both internal and external security assessments of AI systems prior to their release. Investment in cybersecurity to safeguard sensitive AI data was also a key component, as was the development of techniques for watermarking content generated by AI.

Non-Binding Nature and Policy Shift

It’s important to understand that Anthropic had already been implementing many of the practices detailed in these commitments. The agreement itself was not legally enforceable. The Biden administration intended these commitments to highlight its AI policy priorities in anticipation of a more comprehensive AI Executive Order, which was subsequently enacted.

A shift in governmental approach is now evident, as the Trump administration has signaled its intention to pursue a different path regarding AI governance.

Reversal of Biden's Executive Order

In January, President Trump rescinded the previously mentioned AI Executive Order. This order had directed the National Institute of Standards and Technology to create guidelines assisting companies in identifying and rectifying flaws in AI models, including inherent biases.

Critics aligned with Trump contended that the order’s reporting requirements were excessively burdensome and compelled companies to reveal proprietary trade secrets.

Trump's New Directive

Following the revocation of the AI Executive Order, Trump issued a new order instructing federal agencies to foster the development of AI “free from ideological bias.” The goal is to promote “human flourishing, economic competitiveness, and national security.” Significantly, Trump’s order did not address the issue of combating AI discrimination, a central focus of Biden’s initiative.

Commitment Consistency

As highlighted by The Midas Project on social media, the Biden-era commitments contained no stipulations regarding a time limit or dependence on the political affiliation of the current president.

In November, several AI companies affirmed that their commitments remained unchanged following the election.

Policy Adjustments by OpenAI

Anthropic is not the sole company to modify its publicly stated policies since Trump’s inauguration. OpenAI recently announced its commitment to embracing “intellectual freedom… no matter how challenging or controversial a topic may be.” They also pledged to ensure their AI does not censor specific viewpoints.

OpenAI also removed a webpage that previously detailed the startup’s dedication to diversity, equity, and inclusion (DEI). These programs have faced scrutiny from the Trump administration, prompting numerous companies to either eliminate or significantly revise their DEI efforts.

Allegations of AI Censorship

Several of Trump’s AI advisors from Silicon Valley, including Marc Andreessen, David Sacks, and Elon Musk, have asserted that companies like Google and OpenAI have engaged in AI censorship by restricting the responses of their AI chatbots.

Labs, including OpenAI, have denied that their policy changes are a response to political pressure.

Government Contracts

Both OpenAI and Anthropic are currently pursuing or have secured government contracts.

Anthropic's Response

Several hours after the initial publication of this report, Anthropic provided TechCrunch with the following statement:

“We remain committed to the voluntary AI commitments established under the Biden Administration. This progress and specific actions continue to be reflected in [our] transparency center within the content. To prevent further confusion, we will add a section directly citing where our progress aligns.”

Updated 11:25 a.m. Pacific: Added a statement from Anthropic.