- Anthropic faces backlash from users who fear that watermarking will degrade Claude's outputs and even place a 'scarlet letter' on works users create with the chatbot.
- On Friday, Anthropic published a blog post explaining how the watermarking will work and its limitations, making it far from a flawless method for flagging Claude-generated text.
- Anthropic confirms using a watermarking system based on Google's approach to leave an invisible pattern in Claude's responses.
- Watermarking is applied globally by Anthropic to comply with the EU AI Act, despite its limitations.
- Pope Leo XIV releases his first encyclical, 'Magnifica Humanitas', addressing the dangers of AI and making six central claims.
- Chris Olah of Anthropic stands alongside the pope at the encyclical's unveiling, noting that AI labs operate inside conflicting incentives.
- In 'Magnifica Humanitas', Pope Leo XIV warns of AI's private origins, stating that the main drivers of development are often private, transnational parties with resources surpassing many governments.
- The encyclical emphasizes the need for a diversity of views on AI, warning against a corporate monoculture that sacrifices the weak.
- Pope Leo XIV argues that deliberation is preferable to speed in the context of AI, highlighting the importance of education in understanding its use.
- The encyclical calls for new binding principles of justice to prevent exclusion and deprivation of freedoms due to AI technologies.
- Pope Leo XIV links current AI practices to historical injustices like slavery and colonialism, urging vigilance against repeating past mistakes.
- The encyclical warns against the use of AI in weaponry, stating that moral judgment cannot be reduced to calculation.
Anthropic's watermarking system aims to comply with the EU AI Act but has limitations, including concerns over text quality and the inability to confirm authorship. The company states, “In internal testing, we’ve seen no impact of watermarking on the content, level of creativity, or readability of Claude’s text.”1234
However, users fear it may degrade outputs and label their work. The watermarking can’t be applied to small text samples or specific code fragments, as fewer word choices limit its effectiveness.
In a broader context, Pope Leo XIV's encyclical, “Magnifica Humanitas,” warns that AI technologies could exacerbate historical injustices, including slavery. He emphasizes the need for ethical oversight, stating, “Justice demands that we prevent the emergence of new forms of exclusion and deprivation of freedoms.”5
Leo argues that AI's development should not replicate past power imbalances, urging a diverse approach to technology that avoids a corporate monoculture. He highlights the risk of dehumanization, stating, “The risk of dehumanization... is an ancient and ever-new temptation that today takes on a technical guise.”
The encyclical calls for vigilance against the potential for AI to create “second-class” human beings, echoing the historical legacy of slavery and colonialism, and advocating for justice and fraternity in technological advancement.
“Anthropic's watermarking can't trace text to a specific person or chat, and it fails on small samples or code fragments. The pope's encyclical 'Magnifica Humanitas' makes six claims, including that AI in weaponry lowers the moral threshold of conflict.”
