Last Week’s AI News #29

This week is not about tools.

It’s about power shifts.

It is now shaping policy, rewriting industries, and quietly redrawing the boundaries of how businesses operate.

Here’s what you need to know from last week in AI:

• OpenAI proposes taxes, wealth funds, and a new social contract for AI
• Investigation questions Sam Altman’s leadership and past actions
• Anthropic unveils powerful but unreleased “Project Glasswing” system
• Anthropic locks in massive compute deal amid surging growth
• Meta launches Muse Spark, entering the frontier AI race again
• Anthropic simplifies agent building with Managed Agents
• Perplexity turns into a personal finance AI hub
• Oxford researchers detect heart failure risk years in advance with AI
• Everything else that happened in AI last week

OPENAI PROPOSES TAXES, WEALTH FUNDS, AND A NEW SOCIAL CONTRACT FOR AI

OpenAI published a 13-page policy document outlining how society should prepare for the rise of superintelligence. The proposal includes ideas like taxing AI-driven profits, implementing a four-day workweek, and guaranteeing universal access to AI tools.

One of the most striking suggestions is a sovereign-style wealth fund funded by AI companies, distributing dividends to citizens, similar to Alaska’s oil revenue model. The document frames the current moment as the beginning of a transition toward superintelligence, requiring a completely new social contract.

Axios described it as the most detailed blueprint any tech leader has ever released for regulating and redistributing wealth from their own technology.

Why does it matter for businesses?
The CEO of one of the world’s most valuable AI companies is actively proposing systems to redistribute the value his own technology will create. That signals a future where AI doesn’t just disrupt markets, it forces structural economic change, and businesses need to be ready for both opportunity and regulation.

INVESTIGATION QUESTIONS SAM ALTMAN’S LEADERSHIP AND PAST ACTIONS

A major investigation published by The New Yorker examined Sam Altman’s career, drawing from over 100 interviews, internal memos, and private notes from key figures in the AI industry.

The report alleges a long-standing pattern of misleading behavior, including claims that safety protocols were misrepresented to OpenAI’s board. Notes from both Ilya Sutskever and Dario Amodei independently point to concerns about leadership at the company’s highest level.

While no definitive proof of wrongdoing was presented, the report paints a detailed and consistent picture of internal tension and distrust.

Why does it matter for businesses?
Leadership risk at the top of the most influential AI company in the world has ripple effects across the entire ecosystem. Whether the claims hold or not, the situation highlights how much the future of AI depends not just on technology, but on the people building it.

ANTHROPIC UNVEILS POWERFUL BUT UNRELEASED “PROJECT GLASSWING” SYSTEM

Anthropic introduced Project Glasswing, a cybersecurity-focused initiative built around Claude Mythos Preview, an unreleased frontier AI system with extremely advanced capabilities.

The model reportedly identified thousands of vulnerabilities across major operating systems and browsers, including issues that had gone undetected for decades. Benchmarks show strong improvements across coding, reasoning, and other domains compared to existing frontier models.

Despite its performance, the model will not be released publicly. Access is limited to a small group of partners, with significant resources allocated toward defensive cybersecurity efforts.

Why does it matter for businesses?
The most powerful AI systems are increasingly being kept behind closed doors. This signals a shift where cutting-edge capabilities are first deployed in controlled environments, giving early access advantages to select partners while the broader market waits.

ANTHROPIC LOCKS IN MASSIVE COMPUTE DEAL AMID SURGING GROWTH

Anthropic signed a multi-gigawatt agreement with Google and Broadcom, securing 3.5GW of TPU capacity starting in 2027. The deal comes as the company’s revenue growth accelerates rapidly.

Since the start of the year, Anthropic’s run-rate revenue reportedly tripled to $30B, while its base of large enterprise customers doubled. The expansion is aimed at meeting increasing demand and supporting future large-scale models.

The growth is happening despite regulatory pressure, including concerns raised by the U.S. government about supply chain risks.

Why does it matter for businesses?
Demand for advanced AI systems continues to outpace available infrastructure. Companies that secure compute early will have a major advantage, while others may face constraints not from ideas, but from access to resources.

META LAUNCHES MUSE SPARK, RE-ENTERING THE FRONTIER AI RACE

Meta’s Superintelligence Labs introduced Muse Spark, a new multimodal reasoning model capable of handling text, voice, and image inputs.

The system includes a “contemplation mode” where multiple AI agents debate and evaluate complex problems. Early benchmarks show competitive reasoning performance, though the model still lags in coding tasks compared to top competitors.

Unlike previous open models from Meta, Muse Spark is proprietary, signaling a shift in strategy.

Why does it matter for businesses?
Meta is repositioning itself in the AI race with a more focused and competitive approach. While not yet leading, its scale, data, and distribution give it the potential to quickly close the gap.

ANTHROPIC SIMPLIFIES AGENT BUILDING WITH MANAGED AGENTS

Anthropic launched a public beta of Claude Managed Agents, a platform designed to simplify the process of building and deploying AI agents.

The system handles infrastructure, security, and execution, allowing users to go from idea to working product in a matter of days. Agents can operate independently for extended periods and coordinate with other agents on complex tasks.

Early adopters include major companies experimenting with deploying agents across multiple departments.

Why does it matter for businesses?
Building AI agents is moving from an engineering challenge to a product decision. Lowering technical barriers means more companies can deploy automation faster, increasing competition and accelerating adoption across industries.

PERPLEXITY TURNS INTO A PERSONAL FINANCE AI HUB

Perplexity introduced a new integration that connects financial accounts directly to its AI agent, transforming it into a centralized personal finance tool.

Users can link bank accounts, credit cards, and loans, allowing the system to generate budgets, track net worth, and build financial plans through simple prompts. The feature builds on earlier automation capabilities like tax form generation.

The company’s rapid growth reflects strong demand for these agent-driven use cases.

Why does it matter for businesses?
AI products are expanding beyond their original categories. What started as a search competitor is now entering financial services, showing how quickly AI can collapse boundaries between entire industries.

AI DETECTS HEART FAILURE RISK YEARS IN ADVANCE

Researchers at the University of Oxford developed an AI system capable of identifying early signs of heart failure from standard CT scans.

The system detects subtle changes in fat tissue around the heart, revealing patterns that are invisible to human doctors. In high-risk cases, the model was able to predict heart failure years in advance with high accuracy.

The technology is already moving toward real-world deployment in healthcare systems.

Why does it matter for businesses?
AI’s impact is extending into critical, real-world outcomes. Moving from detection to prediction in healthcare doesn’t just improve efficiency, it fundamentally changes how and when decisions are made.

EVERYTHING ELSE THAT HAPPENED IN AI LAST WEEK

• OpenAI healthcare usage growth: ChatGPT is now handling around 2M insurance-related messages weekly, showing rapid adoption in healthcare-related workflows. This signals growing trust in AI for sensitive, real-world use cases.
• Google on-device dictation app: Google released AI Edge Eloquent, an iOS app that converts raw speech into polished text entirely on-device. This enables faster content creation without relying on cloud processing.
• Legion Health AI prescription approvals: Legion Health received approval for its AI app to refill psychiatric medications without clinician oversight. This opens the door for automated healthcare operations at scale.
• Microsoft open-source embedding model: Microsoft open-sourced Harrier, a state-of-the-art embedding model supporting 100+ languages for search and retrieval. This enables businesses to build more accurate AI-powered search systems.
• Perplexity rapid revenue growth: Perplexity reached $450M in ARR after a 50% monthly increase, driven by its agent-based system and usage pricing. This highlights strong demand for practical AI tools.
• Canva AI workspace expansion: Canva acquired Simtheory and Ortto to add agent-based workflows and marketing automation. This enables end-to-end campaign creation inside a single platform.
• OpenAI cybersecurity AI model: OpenAI developed a new model with advanced cybersecurity capabilities, planning limited release to select partners. This signals growing focus on defensive AI systems.
• OpenAI  pro subscription tier: OpenAI launched a $100/month Pro plan with significantly higher usage limits for coding agents. This targets power users building more advanced AI-driven workflows.
• OpenAI, Google, Anthropic model protection efforts: Major AI labs are collaborating to prevent competitors from distilling their models. This reflects increasing competition and protection of AI intellectual property.
• Florida AG OpenAI investigation: Florida’s attorney general opened a probe into OpenAI over concerns about ChatGPT misuse. This highlights rising regulatory scrutiny around AI deployment.

AI is no longer asking for permission.

It’s already inside workflows, infrastructure, and decision-making systems.

The only question left is how fast businesses adapt.

👉 Subscribe to our newsletter to get the next week of AI, before everyone else does.