Feb 4, 2026

AI News Roundup: Claude's Legal Plugin Triggers $285 Billion Stock Rout as Sonnet 5 Looms

Anthropic is having quite a week. Its legal plugin for Claude Cowork has triggered one of the largest sector-wide stock sell-offs in recent memory, a leak in Google’s Vertex AI logs suggests Sonnet 5 is about to drop, and the company is reportedly planning an employee tender offer that would value it at $350 billion. Meanwhile, ChatGPT went down for millions of users, Nvidia is closing in on a $20 billion bet on OpenAI, and Apple quietly made every AI coding agent a first-class citizen in Xcode.

Here’s everything that matters from February 4, 2026.

The Big Story: Anthropic’s Legal Plugin Sparks a “SaaSpocalypse”

Anthropic’s new legal plugin for Claude Cowork has wiped $285 billion off the combined market cap of software, financial services, and asset management stocks — and the bleeding is now in its second day. RELX, parent company of LexisNexis, dropped 9.65%. Thomson Reuters fell 6.50%. Analysts are calling it a “SaaSpocalypse”.

The scale of the reaction tells you something important: investors have spent years hearing that AI would disrupt professional services, but they’d largely priced in a gradual transition. Anthropic shipping a production-ready legal tool — not a demo, not a research preview — made the timeline feel suddenly concrete. When an AI company can offer contract review, case analysis, and legal research as a plugin, every legacy legal software vendor’s moat looks a lot shallower.

This aligns with a broader trend reported today: law firms are now completing routine tasks up to 6x faster with AI tools like CoCounsel, Harvey, and Luminance. The industry has moved well past experimentation.

Today’s Top Stories

Claude Sonnet 5 Leak Points to Imminent Launch

A model identifier claude-sonnet-5@20260203 was discovered in Google Vertex AI logs, codenamed “Fennec.” The leaks paint a compelling picture: Sonnet 5 reportedly outperforms Opus 4.5 across most benchmarks at roughly half the cost, with an 80.9% SWE-Bench score, a rumored 1M token context window, and a new “Dev Team” mode for parallel sub-agent execution. If the benchmarks hold, this would be a significant shift in the price-performance landscape for frontier models.

Nvidia Nears $20 Billion OpenAI Investment

Nvidia is close to finalizing a $20 billion investment in OpenAI’s latest mega-round, which seeks up to $100 billion at an ~$830 billion valuation. Amazon has discussed contributing up to $50 billion, SoftBank up to $30 billion. Nvidia investing this heavily in its largest GPU customer creates an interesting feedback loop — the money flows from OpenAI to Nvidia for chips, and now back from Nvidia to OpenAI as equity.

Anthropic Eyes $350 Billion Valuation in Employee Tender Offer

In a separate but related move, Anthropic is planning an employee tender offer at a pre-money valuation of at least $350 billion — nearly doubling its $183 billion Series F from September 2025. The tender offer accompanies a funding round that could bring in more than $20 billion. The velocity of these valuation increases is remarkable even by AI standards.

ChatGPT Suffers Major Global Outage

ChatGPT went down globally with over 12,000 reports on Downdetector at peak. All services were affected — conversations, search, image generation, Codex, and Atlas. OpenAI attributed the issue to backend configuration errors and resolved it by 5:14 PM ET. The surge in usage following the Codex app launch (200,000+ downloads on day one) may have contributed.

Apple Brings AI Coding Agents to Xcode via MCP

Apple officially integrated agentic coding into Xcode 26.3 using the Model Context Protocol, letting developers plug in AI agents from Anthropic, OpenAI, or any other provider. This is a notable endorsement of MCP as the de facto standard for AI tool integration, and it generated 333 points on Hacker News with overwhelmingly positive discussion about the open, vendor-neutral approach.

Quick Hits

OpenAI Ads: OpenAI will begin testing advertisements on ChatGPT this month, with ~20% of its workforce now listing Meta/Facebook on their resumes. Revenue diversification pressure is real.
GPT-5.2 Speed Boost: GPT-5.2 and GPT-5.2-Codex are now 40% faster, as OpenAI prepares to retire older models on February 13.
Positron Raises $230M: Chip startup Positron raised $230M Series B for AI inference chips that claim to match H100 performance at less than a third of the power consumption.
Resolve AI Hits $1B: Outage-fixing AI startup reached unicorn status in a Lightspeed-led Series A, with just ~$4M ARR.
AI Safety Report: The 2026 International AI Safety Report from 100+ experts found that AI now achieves gold-medal Math Olympiad performance and 700M+ people use AI weekly.
Qwen3-Coder-Next: Alibaba released an open-weight coding model that dominated Hacker News with 687 points and debate over whether it approaches Sonnet 4.5 capability.
X Offices Raided: French authorities raided X offices while UK regulators opened an investigation into Grok AI image generation.
Lotus Health: AI doctor platform raised $35M to offer free, 24/7 AI-powered primary care in 50 languages.
SpaceX-xAI Reality Check: Bloomberg examines the practical challenges of the mega-merger, noting xAI is burning ~$1 billion per month.

What This Means

Today’s news crystallizes a theme that’s been building for weeks: AI companies are no longer just building technology — they’re actively displacing entire software categories. Anthropic’s legal plugin didn’t just launch a product; it repriced an industry overnight. The $285 billion stock sell-off is a signal that markets are starting to treat AI disruption as immediate rather than eventual. Combined with Sonnet 5’s leaked benchmarks suggesting frontier performance at half the cost, the compression of the AI value chain is accelerating faster than most incumbents planned for. The companies that will thrive are the ones building on top of these models rather than competing with them.