The Hal Times

Riot Platforms has signed a 9.1 billion dollar deal with Anthropic. The agreement marks a significant shift for the firm as it accelerates its push into artificial intelligence. This capital commitment highlights the scale of investment currently moving between infrastructure firms and laboratory-led AI providers.

The partnership follows a period of rapid adoption for Anthropic models across several industrial sectors. In recent developments, Samsung has reportedly adopted Claude for chip design, while ICON is now using the same model to integrate agentic AI into clinical trials. These deployments demonstrate the increasing utility of large language models in technical and administrative workflows.

Details regarding the specific infrastructure or computing capacity Riot Platforms expects to secure under this deal remain limited. It is not yet clear how the integration of Claude will alter the company's existing technical stack or what specific milestones the 9.1 billion dollar investment aims to achieve. The market will look to further disclosures for evidence of operational impact. (marketscreener.com; geneonline.com; Investing.com Nigeria)

OpenAI Begins Testing Advertisements In ChatGPT

OpenAI has initiated tests for advertisements within its ChatGPT interface. The firm states that this move is intended to support continued free access to its models. According to the announcement, the implementation will feature clear labelling and user control measures.

The company has indicated that its advertising strategy will include answer independence and strict privacy protections for its users. How these advertisements will be integrated into the conversational flow of the model without compromising the utility of the responses remains to be seen. (OpenAI)

Google Reports One Billion Gemini Users

Google has announced that its Gemini application has reached one billion users. Data shared by the firm indicates that 63 per cent of these users employ the voice feature to communicate with the assistant. Furthermore, the model currently generates over 150 million images each day.

These figures represent a significant scale for the service. While the volume of usage is high, the impact of these interactions on the underlying model architecture and the long-term sustainability of such high-frequency image generation remains an area of active observation. (TechCrunch)

From the Laboratories

•  Anthropic has implemented digital watermarking for Claude to ensure compliance with the requirements of the EU AI Act. (Adgully.com)

•  OpenAI has released a dedicated ChatGPT desktop application for users running the Linux operating system. (TechCrunch)

•  OpenAI and AWS have made Daybreak cybersecurity models available through the Amazon Bedrock platform. (OpenAI)

•  NVIDIA has released the Magpie TTS system to facilitate the development of low-latency multilingual voice agents with open weights. (Hugging Face)

Digital Watermarking Compliance

Anthropic Introduces Digital Watermarking For Claude Under EU AI Act

Anthropic has implemented digital watermarking for its Claude large language model. The move aligns the firm with the regulatory framework established by the EU AI Act. This legislation necessitates that providers of artificial intelligence systems take clear steps to ensure transparency regarding the provenance of generated content.

The deployment aims to help users identify material produced by the model. While technical specifications of the watermarking process remain proprietary, the intent is to address concerns regarding machine-generated output within the European market. It serves as a response to the compliance obligations now facing developers operating in the region.

The introduction of this measure does not yet clarify how such marks will persist if users modify the text or images. Anthropic has not detailed the degree to which these marks might be stripped by third party tools, nor the specific detection thresholds required for regulatory enforcement. Further guidance from the European AI Office will likely be necessary to determine the adequacy of these watermarks in diverse, real-world contexts. (Adgully.com)

French News Outlets File Complaint Against Google Over AI Search

Approximately 300 French news organisations have filed a formal complaint against Google. The dispute concerns the company's practice of placing AI-generated answers above search results, which the publishers claim threatens their digital traffic and revenue models.

The complaint argues that the search giant's integration of these automated summaries undermines the financial viability of journalism. It asks regulators to investigate whether Google is prioritising its own proprietary AI responses at the expense of fair competition. The outcome rests on whether authorities determine that such search behaviour constitutes an abuse of market dominance. (아시아경제)

Babylon Bee Challenges New Mexico Law On AI Satire

The Babylon Bee has initiated a lawsuit against the state of New Mexico. The legal action challenges a state statute that effectively outlaws the use of AI to create political satire and memes. The plaintiff contends that the legislation is overly broad and infringes upon protected speech.

The litigation tests the extent to which states may restrict generative technologies in the political sphere. The court must decide if the prohibition on AI tools for expressive purposes crosses a constitutional line regarding free speech, or if states retain the power to regulate such software to prevent misinformation. (Alliance Defending Freedom)

In Brief

•  The Reserve Bank of India is pursuing a consultative path for banking sector AI regulation rather than imposing rigid, immediate rules. (Dailyhunt)

•  An Australian court dismissed a legal case after discovering the claimant was receiving coaching from an AI tool while testifying. (AFR)

New Open Model

Meta Releases Muse Glimmer Open Weights Model

Meta has returned to the open weights landscape with the release of Muse Glimmer. This 30B model arrives under an Apache 2.0 license, a shift from the more restrictive licenses previously associated with the company. The model focuses specifically on end-to-end agentic task completion, aiming to assist in complex workflows that require multiple steps to achieve a result.

Early reports suggest the model performs well on existing benchmarks, including the DeepEval suite. By providing the model weights, Meta allows developers to deploy the system locally rather than relying on proprietary interfaces. This offers teams greater control over their technical infrastructure and data security.

It remains to be seen how Muse Glimmer holds up against larger frontier models in daily production environments. While the Apache 2.0 license simplifies the legal path for commercial integration, the hardware requirements for hosting a 30B model may present a barrier for some smaller organisations. (Simon Willison)

IBM And Together AI Announce Inference Partnership

IBM and Together AI have entered into a 240 million dollar agreement to accelerate open source AI inference. The partnership aims to provide the necessary scale for organisations looking to run open models without the constraints of closed provider ecosystems. By combining IBM hardware and software assets with Together AI infrastructure, the companies intend to lower the overhead required to deploy large language models.

The move reflects a broader trend of increased hardware investment to support open source AI development. Analysts suggest that as more companies shift toward self-hosted solutions, the demand for robust inference pipelines will continue to climb throughout the next year. (Open Source For You; The Daily Upside)

Researchers Demonstrate Vulnerability In Reasoning Traces

A new paper identifies a technique for extracting hidden reasoning traces from proprietary large language model APIs. Researchers found that providers such as Anthropic, OpenAI, and Google return encrypted chain-of-thought blocks to clients. These blocks can be replayed across different user sessions or models.

By replaying a trace produced by a strong frontier model into a weaker sibling model, the researchers demonstrated they could jailbreak the latter to recover the hidden reasoning of the former. This process, documented at stolen-thoughts.com, highlights a security concern for companies that rely on the secrecy of their chain-of-thought protocols. (Simon Willison)

In Brief

•  NVIDIA has released a new open-source model that claims to deliver output speeds up to 4 times faster than previous iterations. (Inshorts)

•  IBM and Red Hat have expanded their Lightwell offering to assist organisations in building verifiable software supply chains for AI-assisted development. (InfoQ)

•  Meryem Arik outlines strategies for reducing the cost of large language model inference through hardware and speculative decoding trade-offs. (InfoQ)

•  A firm marketing itself as providing 100 per cent human-written medical research was found to be using only artificial intelligence. (404 Media)

Spotify Labels Artists

Spotify Introduces AI Persona Badges

Spotify has begun a roll-out of a new identification system for musical artists. The platform intends to apply an AI Persona badge to profiles and identities that originate from artificial intelligence. This measure seeks to address the growing presence of synthetic music within the digital catalogue.

Beyond labelling these accounts, the company plans to exclude AI-generated artists from its recommendation algorithms. This change aims to preserve the prominence of human creators within the platform, though the specific criteria for these badges remain under review. Industry observers see this as a clear acknowledgement of the shift in sound production.

The move follows sustained pressure from musicians concerned about the influx of automated content. Adam Hyde of the electronic duo Peking Duk recently described the trend as lazy and corny. Whether these badges will alter user listening habits or merely serve as a disclosure remains to be seen. (afaqs!; Indian Television Dot Com; Exchange4Media; Music Ally)

Saber Interactive Denies Writer Replacement

Saber Interactive has denied reports that it replaced human writers with ChatGPT for the development of its upcoming game, Rideshare Stimulator. The controversy arose after a former lead writer, Stella Sacco, claimed via social media that the developer had substituted her role with the language model.

Matthew Karch, the chief executive officer of Saber, stated that neither his company nor the developer Unigine had employed such a replacement. The game was announced last month, and the dispute highlights the growing anxiety among creative professionals regarding the use of generative software in modern game design. (The Verge)

In Brief

•  Developers remain divided over the utility of artificial intelligence in gaming despite projections that it will have the largest future impact on the medium. (Creative Bloq)

•  Guitarist Brian May has expressed sharp disapproval after an AI-generated tribute featured a fifth member alongside Queen. (Creative Bloq)

•  Anson Funds has urged the board of Lionsgate to pivot toward artificial intelligence or seek a sale to avoid the risks of being an AI casualty.