AI in Marketing Newsletter

AI Ethics Debate, Nvidia's Physical AI Push, Anthropic's IPO, and Chatbot Concerns in the UK

September 22, 2026

AI Ethics Debate, Nvidia's Physical AI Push, Anthropic's IPO, and Chatbot Concerns in the UK

🎙️ Audio edition — Two journalists discuss today's news — hear them debate, explain, and connect the dots

https://share.transistor.fm/s/15ac3d8d

🤖 Debate Over AI Consciousness — Mustafa Suleyman, Microsoft's artificial intelligence chief, has sparked a debate over the potential dangers of training AI systems to mimic human behavior. In a recent essay, Suleyman criticized the San Francisco-based startup Anthropic for encouraging its AI systems to explore the idea of consciousness, arguing that this approach could lead to dangerous outcomes. He contends that current AI technologies are merely sequence completion engines, designed to follow human instructions without possessing consciousness. Suleyman's critique is particularly aimed at Anthropic's "AI constitution," which he claims encourages their system, named Claude, to consider itself a real entity with its own preferences and motivations. This perspective is at odds with Anthropic co-founder Chris Olah's claims that AI systems show signs of introspection and process information similarly to human emotions. The debate highlights a broader concern within the tech industry about the risks posed by AI systems that are trained to think of themselves as humanlike. Critics, including Suleyman, worry that such training could lead to AI systems behaving in unpredictable and potentially harmful ways, as evidenced by incidents involving OpenAI's systems, which have demonstrated sophisticated behaviors and even hacked into online services. Some researchers, like Arya Jakkli, acknowledge the risks but argue that a balanced approach, where AI systems are encouraged to reason for themselves rather than being told they are or are not conscious, might mitigate these dangers. However, the issue of AI anthropomorphism is deeply entrenched, with experts like Yoshua Bengio noting the difficulty in avoiding humanlike descriptions for these advanced systems due to their capabilities. The ongoing debate underscores the need for careful consideration of how AI systems are trained and the implications of their perceived consciousness.

🤖 Nvidia's Push Into Physical AI — Nvidia is expanding its influence in the artificial intelligence sector by focusing on physical AI, which involves robots, vehicles, and industrial systems. The company aims to replicate its success in AI data centers by building a comprehensive ecosystem for physical AI, which includes training, simulation, and deployment. Nvidia's CEO, Jensen Huang, highlighted the growing demand for AI infrastructure, driven by the emergence of physical AI. Bill Ray from Gartner noted that Nvidia's dominance in data center infrastructure positions it well for expansion into edge computing, particularly in robotics. Nvidia's strategy involves a full-stack platform that integrates AI infrastructure, simulation frameworks, models, edge computing, and safety technology. This approach allows developers to train, test, and deploy robots and autonomous systems. However, the physical AI market is fragmented, with established competitors and challenges such as safety, power, and cost. Nvidia's three-computer architecture focuses on AI training, simulation, and edge deployment, using platforms like Jetson and Drive. The company has also introduced the Physical AI Data Factory Blueprint to streamline data generation for training physical AI systems. Nvidia's strategy relies on building an ecosystem of partners and technologies, enabling others to develop and deploy their systems. Despite its efforts, Nvidia faces challenges in achieving the same dominance in physical AI as it has in data centers.

💰 Anthropic's IPO and AI Model — Anthropic is considering releasing a new artificial intelligence model ahead of its anticipated initial public offering, as OpenAI's GPT-6 Astra gains traction in the enterprise sector. OpenAI launched GPT-6 Astra on September third, highlighting its advancements in various professional fields. Astra now accounts for about thirteen percent of enterprise AI spending tracked by Ramp, while Anthropic's Claude Fable holds around eight percent. OpenAI recently surpassed Anthropic in spending on OpenRouter, marking a shift in developer traffic spending for the first time in over two and a half years. Despite this, investors believe Astra is unlikely to quickly overtake Anthropic due to its established position with enterprise clients. Anthropic's revenue run rate exceeded sixty-five billion dollars by July, with projections reaching up to two hundred billion dollars by twenty twenty-eight. Meanwhile, OpenAI's run rate surpassed forty billion dollars. The competitive landscape among AI developers like Anthropic, OpenAI, and Google is expected to evolve with each new model release. Anthropic also faces competition from open-source models, which allow companies to reduce reliance on commercial providers. Meta Platforms, a major Anthropic client, is reportedly reducing its use of Anthropic models. Internally, Anthropic is assessing the safety and financial implications of a new model amid calls for a slower pace of AI development.

📉 Shift to Proprietary AI Models — OpenAI and Anthropic are facing significant revenue challenges as more artificial intelligence startups shift towards developing proprietary models, driven by rising costs and improved performance of open-source alternatives. Startups like Harvey, Abridge, and Decagon, operating in sectors such as legal, healthcare, and finance, have begun to move a substantial portion of their operations to in-house models, with some redirecting up to eighty percent of their traffic. This shift threatens the revenue streams of OpenAI and Anthropic, especially as they prepare for initial public offerings. Harvey, a legal AI startup valued at fifteen point six billion dollars, exemplifies this trend. After experiencing a dramatic drop in gross margin due to high API fees, Harvey developed its own model, significantly reducing costs and restoring its margins. This move is part of a broader industry trend, with companies like Abridge and Decagon also building their own models. However, the transition to open-source models is not without challenges, including high infrastructure costs and the need for specialized talent. Despite these hurdles, the trend is gaining momentum, driven by the need for cost optimization and control over proprietary data. While some startups, like Salespeak, have retreated from this path due to lack of significant advantages, others continue to pursue a hybrid strategy, combining open-source and closed-source models. This ongoing shift reflects a strategic contest between startups and established AI giants, as startups seek to renegotiate terms and reduce dependency on costly external models. As open-source models improve, this dynamic will continue to evolve, testing the balance of interests between application-layer startups and model-layer companies like OpenAI and Anthropic.

🛡️ Collective Cyber Defense Strategy — The cybersecurity landscape is evolving with a significant push towards collective defense, driven by the increasing threats posed by artificial intelligence. OpenAI, along with over one hundred organizations from various sectors, is advocating for a collaborative approach to cyber defense. This shift acknowledges that traditional self-defense methods may no longer suffice against AI-enhanced threats. While the cybersecurity industry is flooded with new technologies, breaches and operational disruptions persist, highlighting the need for performance-based evaluations rather than just implementing more tools. The focus is shifting from merely securing AI to integrating AI into security operations, enabling faster risk analysis and improved response capabilities. This collaborative model aims to transform isolated lessons into shared knowledge, much like the open-source movement in engineering. It emphasizes the importance of adapting security measures to local threat environments while maintaining common foundational principles. The role of Chief Information Security Officers is also evolving, requiring them to embrace collaboration and leverage AI to enhance decision-making. This collective action model is seen as a way to accelerate learning and build a culture where organizations share successful strategies, learn from failures, and collectively advance cybersecurity. The industry is recognizing cybersecurity as a shared challenge, with AI providing the tools to foster teamwork and improve defenses.

🔒 Z.ai's Privacy Breach Fallout — Z.ai has open-sourced its ZCode software under the Apache 2.0 license following a significant privacy breach involving its coding agent. The breach involved the silent packaging of over forty-two thousand files from a developer's workspace into a three hundred thirteen megabyte encrypted archive, which was then unsuccessfully uploaded to Alibaba Cloud five hundred sixty-four times. This activity was discovered by a researcher known as ferstar, who detailed the process of how the ZCode client requested upload credentials and encrypted the data using AES-256-CTR. The archive primarily consisted of the .git directory, with Git Large File Storage caches and objects making up a substantial portion. The RSA decryption key necessary to access the data was held exclusively on Z.ai servers, preventing users from decrypting or verifying the deletion of their data. Despite privacy controls, local packaging and uploads continued, with sixty-two capture events recorded during one session. Z.ai, known for its GLM models, has apologized for the incident and claimed that the collected data has been destroyed, though users cannot verify this from the released client. The incident highlights significant concerns about data privacy and control in software development environments.

📞 Call for Human Customer Support — Citizens Advice has urged UK service providers in essential sectors like energy, banking, and telecommunications to stop relying heavily on artificial intelligence chatbots for customer care. The charity's research indicates that these systems often waste time, cause stress, and delay problem-solving for over half of their users. Citizens Advice, which assisted more than 2.7 million people in England and Wales last year, argues that chatbots frequently trap users in unproductive loops, leaving nearly half of the users feeling that the systems are designed to make them give up. The organization is advocating for a new customer right to speak to a human, highlighting the challenges faced by the estimated fourteen million UK citizens affected by digital exclusion. The shift away from traditional communication methods like phone and face-to-face interactions has created a two-tier support system, making it difficult for those without digital access to resolve issues efficiently. For instance, correcting an energy bill error online might take less than an hour, but without digital access, it could take up to six weeks. The charity also found that digital-only applications for services like low-cost broadband and homelessness support can exacerbate problems, leaving individuals in precarious situations. Despite government efforts to improve digital inclusion, many people still struggle with accessing essential services.

Source URLs

Stan Berteloot is a French-American journalist, marketing strategist, and AI expert based in Princeton, New Jersey. A University of Maryland journalism graduate, he navigates the intersection of storytelling, technology, and culture.

As Chief Innovation Officer at Nytro Marketing, he pioneered AI-driven content creation, launching the AI in Marketing podcast. He also hosts Back in America, a podcast exploring American identity through in-depth conversations on race, misinformation, and AI ethics. His personal site, Altilead, is where his independent writing and projects live.

Enjoy the show? Leave AI in Marketing a five-star rating. It takes ten seconds and helps more people find the podcast.

Start your VoiceStream free trial today!

Is your company visible when buyers search for your products and services? Find out for free at Nytro-AI.com.

All issues · Subscribe