OpenAI is pitching investors on the idea that its early infrastructure buildout gives it a decisive advantage over Anthropic. Meanwhile, the company is pausing its UK data center project, and Anthropic is exploring custom AI chips.
According to Axios, OpenAI is working on a new cybersecurity product that will only be available to a small group of companies.
Axios initially reported that OpenAI was releasing a new model, drawing comparisons to Anthropic, which on Tuesday restricted access to its new Mythos Preview model to select technology and security firms because of its advanced hacking capabilities.
Axios has since corrected its reporting: the limited rollout only applies to the cybersecurity product, not OpenAI's upcoming "Spud" model.
The product will be distributed through "Trusted Access for Cyber," a pilot program OpenAI launched in February alongside the release of GPT-5.3-Codex. Participants in the program get access to especially capable models for defensive security work, backed by $10 million in API credits.
Claude Cowork is now available on all paid plans for macOS and Windows. Anthropic is also rolling out organizational controls: role-based access, per-team budget limits, usage analytics, and OpenTelemetry monitoring. A new Zoom connector pulls meeting summaries and tasks directly into Cowork, and admins can restrict specific connector actions like write access.
Claude Cowork is essentially the non-developer version of Claude Code, Anthropic's AI tool that's become popular with programmers. According to Anthropic, knowledge workers in marketing, finance, and law are using the assistant for things like project reports, presentations, and research.
The key difference between Cowork and Claude Chat on the web is that it can access files directly on your local hard drive. The desktop app is available at claude.com/download. Like all agentic systems, it's vulnerable to new cybersecurity risks like prompt injections.
Anthropic partner Microsoft has adapted the Cowork technology for Microsoft Copilot. A version is currently in testing and expected to roll out more broadly soon.
OpenAI is restructuring its subscription tiers with a new $100-per-month Pro plan. The biggest change: significantly more Codex usage for heavy users at half the price of the old Pro tier.
Google Gemini can now turn questions and complex concepts into interactive visualizations right inside the chat. Users can tweak variables, rotate 3D models, and explore data on the fly. The feature is designed to help people dig deeper into content, the company says.
To try it out, head to gemini.google in your browser and select the "Pro" model. From there, phrases like "show me" or "help me visualize" will prompt Gemini to generate a visualization of whatever topic you're looking at. Each visualization can be customized to fit your needs.
Anthropic already rolled out a similar feature for its chatbot Claude back in mid-March. Claude also generates interactive diagrams and graphics directly in the chat when the model thinks it makes sense or when the user specifically asks for it.
A US appeals court has declined to temporarily block the Pentagon's designation of Anthropic as a national security risk, Reuters reports. The ruling came Wednesday in Washington, D.C. Defense Secretary Pete Hegseth had placed the AI company on a blacklist after Anthropic refused to lift usage restrictions on its AI assistant Claude for surveillance and autonomous weapons.
Anthropic calls the move retaliation for its stance on AI safety and warns of billions in damages. The Justice Department says the decision was based on contract terms.
A California court had ruled in Anthropic's favor in a parallel case in late March. It marks the first time a US company has been publicly designated as a supply chain risk. A final ruling is still pending.
Zhipu AI has released its new GLM-5.1 model under an MIT license. The model can reportedly refine its own approach over hundreds of iterations when tackling coding tasks.
Anthropic’s new “Claude Managed Agents” gives developers a hosted platform for building and running autonomous AI agents. Early adopters like Notion and Rakuten are already using the system.
