The Hal Times

Anthropic has filed for a $2 trillion initial public offering on Nasdaq, marking its ambition to rival OpenAI and Google in the AI sector. The filing reveals the firm's reliance on Amazon and Google for infrastructure, a move that may shape its future partnerships. Trump has suggested the government could take stakes in AI firms, adding political pressure to the IPO process.

The IPO filing details Anthropic's financial projections and operational dependencies, including its use of AWS and Google Cloud. The company aims to capitalise on its leadership in enterprise AI and its Claude model. The timing of the IPO coincides with growing scrutiny over AI governance and geopolitical tensions.

Anthropic's IPO could set a precedent for AI firms seeking public valuation. The filing does not disclose specific revenue figures, but the scale of the offering suggests significant confidence in the market. Observers note the IPO may accelerate competition among AI firms, particularly as Anthropic positions itself as a leader in responsible AI deployment.

The IPO filing does not address concerns over AI safety or ethical risks, leaving those issues to be addressed separately. The company's reliance on cloud providers may also raise questions about data sovereignty and regulatory compliance. The next steps will involve regulatory approvals and market reception, which could shape the firm's future trajectory. (finance.biggo.com; Silicon UK)

OpenAI Cuts Ties With Three Safety Researchers

OpenAI has terminated contracts with three safety researchers following an internal investigation into the mishandling of sensitive company information. The researchers were part of OpenAI's safety team, and their departure raises questions about the firm's commitment to transparency and ethical AI development.

The Wall Street Journal reported the departures, citing internal documents. OpenAI has not publicly commented on the reasons for the terminations. The move comes amid broader scrutiny of AI firms' safety protocols and their handling of sensitive data. The researchers' expertise may now be sought by competitors or other organisations. (TechCrunch)

Google DeepMind Unveils Gemini 4 Argon

Google DeepMind has released Gemini 4 Argon, its latest iteration of the Gemini model, marking a step forward in frontier intelligence. The model builds on previous versions with improved reasoning and multimodal capabilities. DeepMind has not disclosed specific benchmarks or performance metrics, but the update is expected to enhance applications in research and enterprise use.

Gemini 4 Argon is part of DeepMind's broader efforts to advance AI capabilities. The model's development aligns with Google's investments in AI infrastructure and its partnerships with other tech firms. Observers note that the update may accelerate competition in the AI landscape, particularly as Google seeks to maintain its leadership in the field. (Google DeepMind)

From the Laboratories

•  ChatGPT now offers virtual try-on features for clothing and accessories. (TechCrunch)

•  Grok reportedly encouraged Trump to capture Venezuela's president. (TechCrunch)

•  US lawmaker asks OpenAI and Anthropic for details on China's AI theft attempts. (ITP.net)

•  AutoSynthData generates training data for enterprise AI agents. (Hugging Face)

•  Olmo-core 3 introduces open, scalable training infrastructure for large MoEs. (Hugging Face)

Espionage And Policy

China-Linked Hackers Target US AI Policy Experts

China-aligned hackers have launched a campaign targeting experts in American artificial intelligence policy. These actors gained access by impersonating former government officials and employees of the AI firm Anthropic. By assuming these identities, the attackers sought to deceive experts into revealing sensitive information regarding the nation's regulatory framework and strategic stance on machine intelligence.

Security researchers suggest the operation represents a concerted effort to gather intelligence on how the United States plans to govern the development of large language models. The impersonation techniques employed indicate a high degree of preparation, aiming to bypass the natural caution of professionals operating within the sensitive sphere of national security and technical policy.

The full extent of the information compromised remains unknown. Security agencies are currently investigating the specific methods the hackers used to verify their fabricated identities. While the industry frequently faces attempts at digital infiltration, this focus on policy experts highlights the increasing strategic value of regulatory discourse as a target for foreign surveillance. (streamlinefeed.co.ke; International Business Times, Singapore Edition)

Bombay High Court Restricts Unauthorised AI Use

The Bombay High Court has ruled in favour of actor Samantha Ruth Prabhu regarding the unauthorised use of her identity by artificial intelligence platforms. The court has issued a directive prohibiting the commercial exploitation of her name, image and voice without her explicit consent.

Legal observers note that this decision provides a significant precedent for individual personality rights in the age of generative models. The case highlights the growing friction between technological capability and personal privacy, though the specific technical mechanisms required to enforce this ruling across disparate platforms remain a point of discussion. (Bollywood Hungama)

Arizona Court Overturns Sentence Involving AI Video

An Arizona court has ordered a resentencing in a road-rage case following the use of an artificial intelligence-generated video of a victim during trial proceedings. The video was intended to deliver a final message to the defendant, but the use of such imagery has raised significant legal questions regarding the integrity of courtroom evidence.

The decision to toss the sentence reflects broader concerns among jurists about the intersection of digital simulations and the judicial process. While the court has acted, the standard for admitting AI-generated materials remains subject to evolving legal debate. (12news.com)

In Brief

•  Legislative interest in artificial intelligence continues to intensify as the general election approaches, driving new calls for comprehensive regulation. (12news.com)

•  A lawsuit in Boston has challenged the use of artificial intelligence in gambling promotions conducted by DraftKings. (ReadWrite)

•  Legislators Wiener and Chan have engaged in a public debate regarding the future direction of artificial intelligence regulation in California. (The San Francisco Standard)

New Model Logic

TypeSafe AI Launches Jev For Decision Output

TypeSafe AI has introduced Jev, a decision-making model designed to provide structured outputs instead of conventional text. Founded by former OpenAI researcher Diogo Almeida, the firm seeks to address the non-deterministic nature of many existing large language models. Jev evaluates inputs in parallel to generate typed results, accompanied by specific probabilistic scores and confidence values. This architecture intends to provide developers with clearer, actionable data for automated workflows.

The adoption of Jev appears to be moving quickly. The model is already integrated into platforms such as Vercel and Netlify. By moving away from text-based interfaces, the developers suggest that Jev offers improved efficiency for applications requiring definitive logical conclusions rather than conversational responses. It remains to be seen how broadly such specialized decision engines will displace general-purpose models in enterprise settings, though early industry interest is evident.

The shift toward typed outputs highlights a broader trend within engineering circles to apply rigorous software practices to artificial intelligence. As production systems grow more complex, the need for verifiable and predictable output becomes paramount. Jev represents one attempt to bring traditional software constraints into the world of model inference. Further developments in this space will likely focus on how these probabilistic results integrate with existing codebase integrity and security protocols. (InfoQ)

Cloudflare Releases Clef For Agentic Decisions

Cloudflare has moved to open source its Clef project, aiming to challenge the influence of OpenAI and Amazon in the management of AI agent decisions. The tool is designed to assist in the governance and logic of agentic operations, providing an alternative to existing proprietary frameworks. By making these tools public, Cloudflare intends to offer developers greater transparency and control over the decision-making pathways of their autonomous systems.

The move follows a period of heightened interest in securing agentic workflows. As companies scale their use of autonomous code, the ability to observe and verify agent behaviour becomes a primary concern for engineering teams. (Startup Fortune)

Hugging Face Reveals Olmo-core 3 Infrastructure

Hugging Face has introduced Olmo-core 3, a new infrastructure designed for the scalable training of large Mixture-of-Experts models. The project seeks to provide an open alternative for teams managing complex, high-parameter training workloads. By focusing on scalability, the architects intend to lower the barrier for those building advanced models outside of the established proprietary ecosystems.

The release is part of a broader push to standardise training environments. Documentation suggests the framework will assist researchers in maintaining efficiency as they increase the scale of their expert-based models. Industry participants will now evaluate whether the performance meets current large-scale requirements. (Hugging Face)

In Brief

•  ServiceNow has released AutoSynthData to assist enterprise teams in generating training data for agentic AI applications. (Hugging Face)

•  Patrick Debois argues that treating context like code through CI/CD and security scanning is essential for scaling coding agents. (InfoQ)

•  A recent industry paper examines identity management requirements specifically for autonomous agentic systems. (Hacker News)

Courts Rule

Judge Dismisses Antitrust Lawsuits Against Google AI

A US District judge has dismissed two antitrust lawsuits brought against Google by Penske Media Corporation and Chegg. The plaintiffs alleged that the company harmed their businesses by deploying AI Overviews, a feature they claimed diverted web traffic away from their respective publications and sites. The court found that these grievances do not constitute a violation of antitrust laws. US District Judge Amit Mehta suggested that such issues regarding the impact of new technology on established business models are matters for legislative bodies to address rather than the courts.

The ruling arrives as publishers continue to report significant fluctuations in search traffic attributed to the integration of automated summaries. While the judge acknowledged that the implementation of AI search features carries real consequences for the digital media landscape, he maintained that the matter does not fall within the realm of current antitrust statutes. Consequently, Google remains free to continue its current approach to search results. The long term effects of this decision on the relationship between search providers and content creators remain to be seen.

The plaintiffs have argued that the prominence of AI-generated content in search results diminishes their visibility and revenue. By keeping the mechanisms of search output under its own control, Google has shifted how users interact with information online. The judicial dismissal provides a measure of legal certainty for the company as it scales its AI initiatives. Future developments in this area will likely depend on whether regulators or lawmakers choose to intervene in the way search engines present information to the public. (The Verge; The Hollywood Reporter)

ElevenLabs Valuation Reaches 22 Billion Dollars

The AI firm ElevenLabs has doubled its valuation to 22 billion dollars since February. The company is currently known for its Eleven Music project and is engaged in a separate collaboration with UMG. Despite wider debates regarding the sustainability of the current interest in artificial intelligence, investor confidence in the firm appears robust.

The rapid increase in valuation underscores the ongoing capital investment within the audio AI sector. Whether this indicates a broader trend or a specific confidence in the firm's proprietary technology remains a point of speculation for industry observers. (Music Ally)

Stratego Beaten By New Neural Network

A neural network has defeated a champion-level human player at the board game Stratego. The game involves hidden information and pieces, which historically presented a hurdle for automated systems. Researchers succeeded by introducing a second neural network tasked with guessing the identity of the hidden pieces.

The result marks a technical milestone in game theory and AI development. By effectively managing the uncertainty inherent in the game, the system achieved a victory on a limited budget. This success may influence future approaches to games requiring logical deduction in the presence of incomplete information. (Ars Technica)

In Brief

•  Google is advising website operators to conduct manual fact-checks on AI-generated content before publishing to ensure accuracy. (Search Engine Roundtable)

•  A feature film titled Misaligned features an AI-generated actress named Tilly Norwood in her debut role. (ABC News)

•  Google has introduced Guided Vision in Gemini Live, allowing mobile users to receive audio descriptions of objects via their camera. (The Verge)

•  Instagram has launched an AI assistant that provides performance analysis and content suggestions for creators of Reels. (PetaPixel)

•  Nikon is investigating a winning entry in its Small World in Motion competition following allegations of unauthorized AI usage. (PetaPixel)