10 May 2026: Cloudflare Confirms AI Made 1,100 Jobs Obsolete as Mythos Rewrites Browser Security

A week of hard truths and new possibilities: Cloudflare becomes one of the clearest examples yet of a profitable company shrinking its workforce because of AI, Anthropic’s Mythos security model quietly patches hundreds of Firefox bugs that had been hiding for over a decade, and Spotify opens a door to AI-generated personal podcasts that live right inside your app.

Cloudflare has announced it is cutting around 1,100 people, roughly 20 per cent of its workforce, while simultaneously posting its highest quarterly revenue in the company’s history. The internet security and performance company, which serves millions of websites worldwide, reported first-quarter 2026 revenues of $639.8 million, up 34 per cent year on year. Yet co-founder and chief executive Matthew Prince was emphatic that the cuts were not a response to financial pressure. “Today’s actions are not a cost-cutting exercise,” he wrote in a post accompanying the announcement.

Prince told analysts that the tipping point came last November, when internal use of AI tools began producing dramatic productivity gains across engineering, finance, HR, and marketing teams. “Employees that were two, 10, even 100 times more productive than they had been before,” he said, comparing the change to switching from a manual to an electric screwdriver. Internal AI usage at Cloudflare has reportedly grown by more than 600 per cent in the past three months alone.

The pattern is becoming familiar across the technology sector. Meta, Microsoft, and Amazon have each described similar dynamics in recent quarters: revenues up, headcount down, AI cited as the cause. What makes Cloudflare notable is the directness of Prince’s framing. Rather than describing the cuts as restructuring or strategic pivoting, he stated plainly that certain support roles will simply not be needed as the company operates in what he called the agentic AI era. Cloudflare employs approximately 5,500 people ahead of the cuts, and Prince suggested the total headcount could be higher again by 2027 once the company hires for new roles AI cannot fill.

Anthropic’s Mythos model, which has not yet been released to the public, has transformed how Mozilla approaches the security of its Firefox browser, finding critical vulnerabilities at a scale and speed that has no precedent in the project’s history. Mozilla’s security researchers published a detailed account this week of their experience using Mythos, which was previewed by Anthropic in April alongside a warning that the model had discovered thousands of high-severity software bugs before it could be made widely available.

The numbers are striking. Firefox shipped 423 bug fixes in April 2026, compared to just 31 in April 2025. Among the discoveries were sandbox vulnerabilities, the kind that normally attract a $20,000 bounty from Mozilla’s bug programme, at a rate far exceeding anything the human researcher community has ever produced. One bug had been dormant in the code for 15 years. Mozilla’s distinguished engineer Brian Grinstead described the change as a genuine turning point: “These things are actually just suddenly very good.” He added that AI tools are finding more sandbox issues than human researchers ever did, a striking statement given the financial incentives involved.

Spotify has launched a tool that lets users generate AI-made personal podcasts and listen to them directly in the app, making it the first major streaming platform to build native support for AI-created personal audio content. The feature, currently in beta, works by connecting Spotify to AI coding tools such as OpenAI’s Codex or Anthropic’s Claude Code. Users write a prompt describing the podcast they want, the AI generates the audio, and the resulting episode appears in their Spotify library, visible only to them.

The practical uses are already emerging. The company offered examples such as a deep dive into World Cup history before a tournament, or a summary of class notes the night before an exam. Other apps including Google NotebookLM and Adobe Acrobat have offered similar podcast-generation features, but Spotify’s move puts a distribution layer in place: your AI-made audio lives inside the same app where you already listen to music and shows. Developers can access the integration through a CLI tool published on GitHub. UK users with a Spotify account and access to one of the supported AI tools can try it today.

OpenAI has added three new voice intelligence models to its developer API, including a live translation tool that supports more than 70 input languages and a new transcription service that captures speech as it happens. The flagship addition is GPT-Realtime-2, built on GPT-5-class reasoning and designed to handle more complex conversational requests than its predecessor. The translation model, GPT-Realtime-Translate, covers 13 output languages and is billed per minute. The transcription tool, GPT-Realtime-Whisper, delivers live speech-to-text in real time.

OpenAI is pitching the tools at customer service teams, educators, and media companies, though any developer with API access can use them. For small businesses that currently rely on human interpreters or recorded voice menus, the per-minute billing model lowers the barrier to building a multilingual voice interface considerably. OpenAI says it has embedded guardrails to detect and halt conversations that violate its harmful content policies.

A startup called Basata has raised $24.5 million to solve a problem that will be familiar to anyone who has tried to see a specialist doctor recently: the administrative backlog that means referral letters go unanswered for days or weeks. The company, founded by former executives from Lyft, Cruise, and Medtronic, has built an AI system that reads incoming referrals, extracts the relevant clinical information, and then uses an AI voice agent to call the patient directly and book an appointment. Co-founder Kaled Alhanafi says the goal is for a patient to have a confirmed appointment by the time they reach the car park after seeing their GP.

The problem is acute in the United States but has clear echoes in the UK, where NHS waiting list pressures have made referral delays a recurring political issue. Basata says it has processed referrals for roughly 500,000 patients to date, with 100,000 coming in the past month alone. The business is usage-based, charging per document processed and per call handled rather than per seat. The $21 million Series A was led by Basis Set Ventures, with participation from Cowboy Ventures.

Voice AI startup Wispr Flow has reported accelerating growth in India following the launch of Hinglish support, a blend of Hindi and English that reflects how hundreds of millions of people in the country actually communicate. The development is significant because it illustrates one of the core challenges facing global voice AI products: multilingual markets, where users naturally switch between languages mid-sentence, have proved far harder to serve than English-first markets. Wispr Flow’s approach of training on real code-switching patterns rather than treating languages as separate modes appears to be gaining commercial traction. The company did not disclose specific user numbers.

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

• Gemini API File Search goes multimodal — Google has expanded its Gemini API File Search to support multimodal retrieval-augmented generation, meaning developers can now search across images, audio, and documents in a single query. Google Blog
• LLMorphism: When users start thinking like language models — A new academic paper explores the phenomenon of people adopting AI-like communication patterns after extended use of large language models, raising questions about cognitive adaptation to AI tools. arXiv
• OncoAgent: AI for oncology clinical decisions — Researchers have published details of a dual-tier multi-agent AI framework designed to support cancer treatment decision-making in clinical settings while preserving patient privacy. Hugging Face
• Task paralysis and AI — A widely read blog post this week examines how reliance on AI tools may be contributing to decision fatigue and difficulty starting tasks independently, a pattern some researchers are beginning to study formally. g5t.de
• Moonshot AI raises $2 billion at $20 billion valuation — Chinese AI lab Moonshot AI, maker of the Kimi chatbot, has closed a $2 billion funding round as demand for open-source AI alternatives to Western models continues to grow in Asia and beyond. TechCrunch

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.