Good morning. It's Thursday, May 14, and we're covering a self-improving AI startup, how “vibe coding” is reshaping scientific research, and why AI-driven job replacement is already reaching the courtroom.

Know someone who'd love this? Forward it their way. First time reading? Sign up here to keep up with the future of tech.

Rather than rubber-stamp AI as cause for termination, a Hangzhou court awarded ~260,000 yuan (£28,000) to a quality-assurance supervisor — surnamed Zhou — who was fired after refusing a 40% pay cut when his employer claimed AI could do his job overseeing large language models. State media framed the ruling as protective signaling for the automation era, building on an earlier Beijing arbitration that found AI adoption alone doesn't meet the "significant change in objective circumstances" bar required to terminate an employment contract.

Anthropic now has more verified business customers than OpenAI, according to Ramp expense data tracking over 50,000 companies. Ramp’s May 2026 AI Index shows 34.4% of surveyed businesses paying for Anthropic services, compared with 32.3% for OpenAI, marking Anthropic’s first lead. Anthropic’s share has climbed from 9% a year ago, while OpenAI’s slipped 1%.

Nearly 50,000 Lake Tahoe residents may lose their longtime electricity source after NV Energy redirects capacity toward data centers tied to the AI boom. Liberty Utilities says the Nevada provider will stop supplying most of the region’s power after May 2027, forcing a scramble for replacement energy. Northern Nevada data centers could add 5,900 megawatts of demand by 2033, as California and Nevada regulators clash over who protects ratepayers.

President Trump reportedly travels to China alongside top executives from Tesla, NVIDIA, Apple, BlackRock, Boeing, and Goldman Sachs as Washington seeks new economic agreements with Beijing. The unusually large delegation spans AI chips, finance, aerospace, manufacturing, and commodities, highlighting how deeply U.S. corporate interests remain tied to China despite years of trade tensions.

Higgsfield is the world's first AI-powered streaming platform, dropping original series made entirely by artificial intelligence. No studios, no gatekeepers, no $200M budgets. Just stories that shouldn't exist yet, streaming right now for free.

Stream for Free.

AI is accelerating cyberattacks, from AI-found zero-days to self-spreading malware, sparking fears of an AI-powered hacking era.

AI infrastructure startup Modal is in talks to raise between $150 million and $250 million at a roughly $4.5 billion valuation, according to people familiar with the discussions. The proposed valuation would represent an 80% jump from Modal’s last funding round just months ago, fueled by annualized revenue growth to around $300 million — up fivefold since fall 2025. Much of that demand comes from Modal’s sandbox environments, which developers use to safely run AI agents and coding tools without affecting broader systems or code bases.

Investors including Accel and existing backer Redpoint Ventures are reportedly involved in the talks, as growing adoption of AI agents drives demand for GPU access and infrastructure software. → Continue reading here. (Paywall)

Recursive Superintelligence, a six-month-old startup founded by former OpenAI, Meta, Google, and Salesforce researchers, has raised more than $650 million at a valuation exceeding $4 billion. Led by AI researcher Richard Socher, the company is pursuing “recursive self-improvement” — the idea that AI systems could eventually improve and develop new AI models with minimal human input.

Backers include Google Ventures, Greycroft, NVIDIA, and AMD, while the startup has already recruited prominent figures such as former Google research chief Peter Norvig. The effort reflects a growing industry push toward automating AI research itself, as coding-capable models from OpenAI and Anthropic increasingly reshape software development workflows. Researchers involved say fully self-improving AI remains years away, with humans still central to generating new ideas. → Read the full article here. (Paywall)

Researchers are increasingly using AI coding assistants such as Claude Code, Cursor, GitHub Copilot, and Codex to generate software, analyze data, and build scientific visualizations through conversational prompts — a practice widely known as “vibe coding.” Climate researcher Zeke Hausfather used the technique to create a “thermal helix” animation showing rising global temperatures, despite not knowing how to code the visualization himself.

Scientists interviewed by Nature said AI coding tools can dramatically speed up prototyping and experimentation, but they also warned about bugs, hallucinations, and hidden errors in AI-generated code. Adoption is accelerating across both academia and software engineering: a 2026 DX survey found more than 90% of developers use AI coding assistants monthly, while fully AI-authored code now accounts for over a quarter of customer-facing software. → Read the full article here.

Utah Approves Giant AI Hub: Stratos’ AI hub drew backlash over water use and a 9GW power demand larger than Utah’s current electricity use.

Meta Adds Incognito AI: WhatsApp’s Incognito mode runs Meta AI in a secure enclave the company says even Meta cannot access.

Doctors Quietly Embrace AI: OpenEvidence now aids 65% of U.S. doctors, despite growing concerns over hallucinations and overreliance.

Poppy Wants Your Attention: New AI app merges calendars, messages, and location data into proactive reminders and suggestions.

Brits Turn to AI Doctors: One in seven UK adults now use chatbots for health advice instead of seeing a doctor, a new study found.

Yesterday’s passage was REAL. It was an excerpt from The Fellowship of the Ring by J.R.R. Tolkien. Here’s how you voted:

🟩🟩🟩🟩🟩⬜️ AI (63.5%)

🟨🟨🟨⬜️⬜️⬜️ Real (36.5%)

Tune in next week for another Real or AI test.

Thanks for reading. See you next time!

— Matthew Berman, Nick Wentz & the Forward Future Team