A summary of the week's top AI stories: OpenAI's robot tax proposal, Meta's $14B Muse Spark debut, Anthropic's zero-day discovery, Visa's agentic payment rails, and the enterprise AI resistance nobody's talking about.
The week's most urgent AI stories, decoded.
OpenAI just told governments to tax robots and mandate four-day workweeks. Meta shipped its first post-Llama model built by a $14.3 billion hire. Anthropic found a 27-year-old OpenBSD bug and committed $100M to fix the internet's infrastructure. Visa built the payment rails for an economy where AI agents buy things. And 80% of enterprise workers are quietly refusing to use the AI their bosses are mandating. Eight stories from the week the agentic economy confronted its own contradictions.
OpenAI Wants Robot Taxes and a Four-Day Workweek
The Company Building AGI Now Wants to Tax It.
OpenAI released a 13-page policy document on April 6 titled "Industrial Policy for the Intelligence Age" proposing a robot tax — where AI systems pay the same taxes as the humans they replace — alongside a public wealth fund seeded by AI companies and a 32-hour workweek with no pay reduction. The paper warns that AI-driven growth could hollow out the tax base funding Social Security, Medicaid, SNAP, and housing assistance as corporate profits expand and reliance on labor income shrinks. The timing, weeks before OpenAI's expected IPO, frames Sam Altman as a policymaker — not just a builder.
The company most responsible for accelerating AI displacement is now lobbying to tax the displacement. Cynics will call it PR. Pragmatists will note that no one else in the industry has put forward anything this specific — and the window for proactive policy is closing faster than Congress moves.
Muse Spark Is Alexandr Wang's $14B Debut.
Meta released Muse Spark on April 8 — the first model from Meta Superintelligence Labs, led by former Scale AI CEO Alexandr Wang, whom Meta recruited for $14.3 billion including a 49% stake in Scale AI. Muse Spark is natively multimodal (voice, text, image input) with a "Contemplating mode" that orchestrates multiple agents reasoning in parallel — a direct competitor to Gemini Deep Think and GPT Pro. Meta acknowledges a gap in coding, but the model is competitive on multimodal understanding and health information. Free across Meta AI, Facebook, Instagram, and WhatsApp. The strategy: frontier capabilities at zero cost.
Muse Spark isn't the best model — Meta admits that. But free multimodal reasoning with multi-agent orchestration across 3 billion monthly users is a distribution play no one else can match. The moat isn't capability; it's scale.
Anthropic Built an AI That Finds Zero-Days Faster Than Humans
Then Decided It's Too Dangerous to Release.
Anthropic announced Claude Mythos Preview and Project Glasswing on April 7 — a consortium using its most powerful model to find and patch security vulnerabilities in critical infrastructure. In a thousand scaffold runs costing under $20,000, Mythos discovered a 27-year-old OpenBSD vulnerability that would let attackers remotely crash any machine running it, alongside thousands of high-severity zero-days across every major OS and browser. The capabilities are so potent that Anthropic won't release the model publicly, restricting access to vetted partners. Anthropic is committing $100M in credits and $4M in direct donations to open-source security organizations.
This is the first time an AI lab has withheld a model not because it generates harmful text, but because it's too good at hacking. Anthropic just drew a new line in responsible deployment — and every other lab will be measured against it.
Visa Just Built the Payment Rails for an Agent Economy
Visa unveiled Intelligent Commerce Connect on April 8 — a platform that enables AI agents to browse product inventories, select items, and complete purchases autonomously using Visa and non-Visa cards. Through a single integration via Visa Acceptance Platform, merchants get secure payment initiation, tokenization, spend controls, and authentication for agent-driven transactions. The platform supports four major agent protocols: Trusted Agent Protocol, Machine Payments Protocol, Agentic Commerce Protocol, and Universal Commerce Protocol. Currently in pilot with AWS, Highnote, Mesh, and others. The significance: Visa is positioning itself as the default payment infrastructure for the agentic economy.
When the world's largest payment network builds infrastructure specifically for AI agents to transact, agentic commerce stops being a demo and starts being a market. The companies that integrate first will define how agents spend money — and that's a winner-take-most dynamic.
80% of Workers Are Refusing Their Company's AI Mandates
The Quiet Rebellion Is Louder Than You Think.
A Fortune investigation published April 9 reveals that 80% of enterprise employees are outright refusing AI adoption mandates — while simultaneously using AI tools covertly on their own terms. The phenomenon, driven by "FOBO" (Fear of Becoming Obsolete), has created a paradox: workers adopt AI privately to boost productivity but resist official implementations they see as threats to their roles. Gen Z workers are actively sabotaging company AI rollouts, according to a separate Fortune report from April 8. Meanwhile, a Gusto survey found 45% of workers using AI without informing managers. The resistance is real, measurable, and accelerating.