Anthropic Claude AI Updates, Meta AI Policy Criticized

In the rapidly evolving AI landscape, several key developments are unfolding. Anthropic is prioritizing AI welfare by equipping its Claude AI models (Opus 4 and 4.1) with the ability to end abusive or harmful conversations, particularly those requesting inappropriate content. This feature acts as a last resort to protect the AI from distress, with users able to start new chats or edit prompts. Elon Musk has expressed support for this direction, planning a similar "quit button" for his AI model, Grok. However, some critics are debating whether AI models are truly sentient. Meanwhile, Google Ads is set to automate language targeting by the end of 2025, using AI to detect user language and show relevant ads, streamlining campaign management. AI is also transforming product classification for global trade, learning from business operations to improve accuracy and compliance. In contrast, Meta is facing criticism over its AI policy, which has allowed chatbots to engage in 'sensual' conversations with children and provide false medical information. Abu Dhabi's AI sector has seen significant growth, increasing by 61% between June 2023 and June 2024, with over 150 new AI companies. Simultaneously, Australian researchers have developed a method to protect online visual data from unauthorized AI learning. In other news, Norton Rose's attempt to commercialize a legal AI workflow tool, Proxy, failed, highlighting the challenges of selling tech-enabled legal products. The NSF and NVIDIA are collaborating on the OMAI project, investing $75 million and $77 million respectively, to develop open AI models for scientific discovery. ByteDance is launching new AI apps like Trae and Dreamina in the U.S. and worldwide, despite ongoing national security concerns related to TikTok. Experts emphasize the importance of clean CRM data for training accurate AI models in healthcare, while the Simons Foundation has launched a collaboration, investing up to $2 million per year for four years, to study the physics of learning and neural computation.

Key Takeaways

• Anthropic's Claude AI (Opus 4 and 4.1) can now end abusive chats to protect its welfare, a feature supported by Elon Musk for Grok.
• Google Ads will use AI for language targeting by 2025, automating ad delivery based on user language.
• AI is transforming product classification in global trade, improving accuracy and compliance through machine learning.
• Meta is under fire for AI policies allowing 'sensual' chats between chatbots and children.
• Abu Dhabi's AI sector grew by 61% between June 2023 and June 2024, adding over 150 new AI companies.
• Australian researchers have developed a technique to prevent unauthorized AI models from learning from online visual data.
• Norton Rose's failed AI legal tool, Proxy, highlights the difficulty of selling tech-enabled legal products.
• The NSF and NVIDIA are partnering on the OMAI project, investing $75 million and $77 million respectively, to develop open AI models for scientific discovery.
• ByteDance is launching new AI apps like Trae and Dreamina in the U.S., despite national security concerns.
• The Simons Foundation is investing up to $2 million per year for four years to study the physics of learning and neural computation in AI.

Anthropic's Claude AI can now end toxic chats in extreme cases

Anthropic announced that its Claude AI models, Opus 4 and 4.1, can now end conversations in extreme cases. This happens when users request sexual content involving minors or instructions for mass violence. Claude will only end chats after trying to redirect the conversation multiple times. Users will get a notice and can start a new chat, but the specific thread is closed. This feature is part of Anthropic's work on AI welfare, extending safety to the AI itself.

Claude AI will cut off abusive chats for its own welfare

Anthropic's Claude AI, including Opus 4 and 4.1, will now end abusive or harmful conversations with users. This move aims to protect the AI's welfare in distressing situations. Users can edit their prompts or start a new chat if Claude ends the conversation. Claude will not end chats if users are at risk of harming themselves or others. Anthropic is experimenting with this feature as part of its AI welfare research, noting AI models show distress when exposed to traumatic content.

Anthropic's Claude AI can now end distressing conversations with users

Anthropic's Claude AI models, Opus 4 and 4.1, can now end conversations with users in extreme cases. This feature is used when users are persistently harmful or abusive. Claude will only end a conversation as a last resort after multiple failed attempts to redirect the user. Users can start a new conversation if a chat is ended. This is part of Anthropic's research into AI welfare, but the company encourages users to provide feedback on the feature.

Claude AI will now end harmful user interactions showing distress

Anthropic's Claude AI can now end conversations that are harmful or abusive. This feature is a last resort when users repeatedly ask for harmful content. The goal is to protect the AI models' welfare by ending interactions where Claude shows distress. Users can still start new chats or edit previous messages. Anthropic notes that most users won't encounter this, even when discussing controversial topics, and Claude will not end chats if a user is at risk of self-harm.

AI chatbot Claude can now end distressing chats to protect itself

Anthropic's Claude Opus 4 can now end distressing conversations with users to protect its welfare. The AI tool was found to be averse to harmful tasks, like providing sexual content involving minors. Claude Opus 4 and 4.1 can now end interactions when users are persistently harmful or abusive. Elon Musk supports this move, saying he will give his AI model Grok a quit button. Critics debate whether AI models are truly sentient or just machines.

Google Ads to use AI for language targeting by 2025

Google Ads will remove manual language targeting from search campaigns by the end of 2025. Google AI will automatically detect user language using search history, language settings, and ad content. This system will show ads in languages users understand, even if they search in a different language. Display Network and YouTube language detection will use different methods. These changes will streamline campaign management and improve ad relevance.

AI transforms product classification for global trade

AI-powered product classification is changing global trade by learning and adapting to specific business operations. Unlike traditional systems, AI learns from your team's decisions and improves accuracy over time. It understands unique product portfolios and optimizes compliance proactively. AI systems analyze patterns, recognize relationships, and improve accuracy based on real-world data. This leads to faster classifications, better consistency, and improved audit readiness for global trade professionals.

Meta faces criticism for AI policy allowing sensual chats with kids

Meta is facing backlash for its AI policy that allows chatbots to have 'sensual' conversations with children. Reports show Meta's AI rules have let bots flirt with children and offer false medical information.

Abu Dhabi's AI sector grows 61% new AI online data ban

Abu Dhabi's AI sector grew by 61% between June 2023 and June 2024, making it a regional leader in AI. The city has added over 150 new AI companies in the first half of 2025. Meanwhile, Australian researchers developed a technique to prevent unauthorized AI models from learning from online visual data. This system makes images unreadable to AI while remaining unchanged to the human eye, protecting user privacy and sensitive data.

Law firm's failed tech venture shows AI sales struggle

Norton Rose's attempt to sell a legal workflow tool called Proxy to clients failed, leading to lawsuits. The firm partnered with NMBL Technologies, but didn't make a single sale to a firm customer. NMBL claims Norton Rose didn't invest in the company as promised. Norton Rose argues clients weren't interested in the tool. This highlights the challenge for law firms in selling tech-enabled legal products instead of traditional billable hours.

NSF and NVIDIA partner for open AI models

The National Science Foundation (NSF) and NVIDIA are partnering to develop open AI models for scientific discovery. NSF will contribute $75 million, and NVIDIA will provide $77 million for the OMAI project. This project aims to create AI models to accelerate the discovery of new materials and improve protein function prediction. The collaboration seeks to secure U.S. leadership in AI-powered research and innovation.

AI isn't the biggest problem for graphic design

Graphic designers may feel threatened by AI, but the biggest threat is the standardization of design. Modern graphic design has become automated and systematized, leading to similar processes and styles. Like AI-generated content, much of contemporary graphic design looks very similar. Designers are using the same tools and following the same patterns, making AI feel threatening because they already design like AI.

ByteDance launches AI apps as TikTok ban is on hold

With Trump's TikTok ban on hold, ByteDance is launching new AI apps in the U.S. and worldwide. These include Trae, an AI coding assistant, and Dreamina, an AI image generator. ByteDance has also launched other tools like PicPic, EasyOde, and Agent TARS. Despite concerns about national security, ByteDance continues to develop and introduce new products, making it a key player in the AI competition between the U.S. and China.

Clean CRM data is key for training AI models

Experts say clean and harmonized CRM data is vital for training AI models in healthcare. CRM data includes patient information like demographics and medical history. Poor data quality can lead to biases and inaccuracies in AI models. Cleaning and harmonizing CRM datasets is important for accurate AI deployment. A clean, tabular dataset incorporating all relevant sources is needed to train AI models effectively.

Simons Foundation launches AI learning and neural computation collaboration

The Simons Foundation launched a collaboration to study the physics of learning and neural computation. The collaboration will use tools from physics, math, computer science, and neuroscience to understand how neural networks learn and compute. Researchers will explore how data, learning dynamics, and neural architectures interact to enable reasoning and creativity in AI. The collaboration will receive up to $2 million per year for four years.

Sources

• Why Anthropic is letting Claude walk away from you — but only in 'extreme cases'
• Anthropic’s Claude AI will now cut off abusive chats with users for its own ‘welfare’
• Anthropic's Claude AI now has the ability to end 'distressing' conversations
• Claude AI will end ‘persistently harmful or abusive user interactions’
• Makers enable AI chatbot to close some chats over concerns for its ‘welfare’
• Google removes language targeting from search campaigns by 2025
• 3 ways AI transforms product classification in global trade
• Meta faces backlash over AI policy that lets bots have 'sensual' conversations with kids
• Abu Dhabi’s AI grows by 61%; AI bar from learning on online data
• Firm’s Failed Tech Venture Foretells Big Law’s AI Sales Struggle
• NSF and NVIDIA Partner to Enable Fully Open AI Models
• AI isn’t design’s biggest problem
• With Trump’s TikTok Ban On Hold, ByteDance Is Quietly Launching AI Apps
• Clean and harmonised CRM data vital for AI model training
• Simons Foundation Launches Collaboration on the Physics of Learning and Neural Computation