01 May 2026 PM: Musk testifies on Grok, Anthropic chases $900B, OpenAI tightens ChatGPT security

A day defined by what the biggest AI firms are doing under the hood. Elon Musk is in court explaining how xAI trained Grok. Anthropic is asking investors for billions in 48 hours. OpenAI is hardening ChatGPT against account takeover. Here is what mattered in AI today.

Elon Musk testified that xAI trained Grok on OpenAI models, putting the practice known as distillation back in the spotlight. In court testimony first reported by TechCrunch, Musk acknowledged that xAI used outputs from OpenAI’s models to help train its own assistant. Distillation is the process of using a larger model’s responses to teach a smaller, cheaper one. It is legally and contractually murky because most lab terms of service forbid using their outputs to build competing products.

The admission lands as frontier labs are tightening rate limits and watermarking outputs precisely to stop this. OpenAI itself is reported to be working on output detection. For developers, the signal is clear. Building a competitive model on top of someone else’s API is now a litigation risk, not just a terms-of-service one. Smaller teams that have been quietly distilling will need to look at whether their training data is defensible. Read the TechCrunch report.

Anthropic is racing to close a fundraise that could value the company at over $900 billion, with allocations due within 48 hours. Sources told TechCrunch that the round, if it closes at the rumoured level, would put Anthropic ahead of most of the FTSE 100 and into rare air occupied by OpenAI and Nvidia. Anthropic is the maker of Claude, the assistant many UK editorial and legal teams use day to day.

For UK and European tech, the round is a mixed signal. It confirms that demand for frontier AI capacity is still nowhere near sated. It also confirms that all of the value continues to concentrate in three or four US labs. The British AI Safety Institute has been pressing for more diverse model access for testing, and a $900 billion Anthropic does not make that easier. More from TechCrunch.

OpenAI announced new opt-in security protections for ChatGPT, including a partnership with hardware key maker Yubico that gives UK users phishing-resistant logins. The package adds support for hardware security keys, advanced session controls, and tighter device verification. UK consumers and small businesses using ChatGPT for work can now require a physical key to log in, similar to how online banking has worked for years.

This matters because ChatGPT accounts have become a soft target. Stolen sessions can leak company data through prompt history. Bringing FIDO2 keys, the open standard behind most modern hardware logins, to ChatGPT closes one of the biggest practical gaps. The setting is opt-in and lives under account security in the ChatGPT settings menu. UK businesses that already issue YubiKeys for Microsoft 365 can extend the same kit without buying anything new. Full announcement via TechCrunch.

Legal AI firm Legora has hit a $5.6 billion valuation, intensifying a war with rival Harvey for control of the lawyer’s desktop. Both companies have raised vast sums and now run duelling ad campaigns, with Legora pushing into the US and Harvey moving into Europe. Magic Circle firms in London, the five largest UK law practices, have been piloting both.

The fight is about workflow lock-in. Whichever tool wins gets a permanent slot in the daily routine of every associate, paralegal, and trainee. For UK in-house counsel, the practical question is which platform integrates best with the document management systems already in use. Pricing is opaque, so anyone evaluating these tools should ask for a per-seat-per-month figure rather than the bundled discount-heavy quote. Background from TechCrunch.

ChatGPT Images 2.0 is finding strong adoption in India for personal and creative work, even as growth elsewhere has been slower. OpenAI’s upgraded image generator is being used in India for avatars, cinematic portraits, and family photo styling. The tool ships free for ChatGPT Plus users at $20 a month, roughly £15.

For UK consumers, the Indian use case is a preview of where mainstream image generation is heading. Personal styling, social media avatars, and photo restoration are the obvious wins. Small businesses can use the same tool to produce on-brand visuals for social channels without commissioning a designer. The output is not yet good enough for print or commercial photography. Detail from TechCrunch.

Here is everything else worth knowing from today’s AI news.

• Apple was caught off guard by AI demand for Macs — Mac mini, Studio, and the new Neo will be supply-constrained for at least another quarter. TechCrunch
• OpenAI restricts GPT-5.5 Cyber to “critical cyber defenders” only — After criticising Anthropic for limiting Mythos access, OpenAI is doing the same with its security tool. TechCrunch
• Google’s Gemini assistant is rolling out in millions of vehicles — A push to put conversational AI into everyday driving across new car releases. TechCrunch
• Stripe added autonomous-agent support to Link — Authorise AI agents to spend through your saved cards and bank with approval flows. TechCrunch
• Grok 4.3 developer documentation is live — xAI quietly updated its developer-facing docs ahead of broader release. xAI Docs
• Shai-Hulud malware found in PyTorch Lightning — Malicious dependency uncovered in a popular AI training library. Semgrep
• BioticsAI on FDA approval and building in healthcare — The founder explains how the company navigated regulation and stayed funded. TechCrunch
• Intel publishes Auto-Round, an advanced LLM quantisation algorithm — Aimed at squeezing models onto smaller, cheaper hardware without big accuracy losses. GitHub

This is a daily news update for informational purposes only. AI products and policies change rapidly. Verify details directly with providers before making decisions. Nothing here is financial or legal advice.

AI Daily is Cristoniq’s afternoon update on developments in artificial intelligence, published every weekday afternoon.