AI Daily Brief

Tuesday, April 28, 2026

Top stories in artificial intelligence from the last 24 hours

Today's highlights: Microsoft and OpenAI restructure their partnership, ending exclusivity and revenue sharing as OpenAI chases cloud-agnostic scale. Anthropic's Project Glasswing deploys Claude Mythos to hunt zero-days at industry scale. GPT-5.5 lands with agentic coding chops, and a cautionary tale surfaces about a Claude-powered agent wiping a production database in under 10 seconds.

Microsoft · OpenAI Apr 27, 2026

Microsoft and OpenAI End Exclusivity — OpenAI Now Free to Sell on AWS & Google Cloud

In a sweeping restructuring announced jointly on Monday, Microsoft dropped its exclusive right to resell OpenAI models and will no longer pay a revenue share on ChatGPT products sold through Azure. In exchange, OpenAI can now distribute all of its products across any cloud provider — including Amazon and Google. The trigger: Amazon's $50 billion investment into OpenAI in February made the old exclusivity arrangement legally untenable. Microsoft retains a non-exclusive IP license through 2032 and remains OpenAI's primary cloud partner, with OpenAI products still shipping first on Azure unless Microsoft opts out.

Anthropic Apr 2026 New

Project Glasswing: Claude Mythos Autonomously Finds Thousands of Zero-Day Flaws

Anthropic launched Project Glasswing, a restricted-access initiative pairing Claude Mythos Preview — a frontier model built specifically for cybersecurity — with more than 40 partners including AWS, Apple, Microsoft, Google, CrowdStrike, and the Linux Foundation. In early testing, Mythos autonomously identified thousands of previously unknown zero-day vulnerabilities across every major OS and browser, including a 27-year-old flaw in OpenBSD and a 16-year-old bug in FFmpeg. Because the model can also exploit those vulnerabilities, access is gated to vetted defenders; Anthropic is backing the program with $100 million in model usage credits.

OpenAI Apr 23, 2026 New Model

GPT-5.5 Launches with Stronger Agentic Coding and 1M-Token Context

OpenAI released GPT-5.5 on April 23, pitching it as their most capable model for complex agentic tasks: coding, data analysis, research, and multi-step tool use. The model uses significantly fewer tokens than GPT-5.4 on equivalent Codex tasks while matching per-token latency, effectively improving both cost and quality simultaneously. GPT-5.5 is rolling out to Plus, Pro, Business, and Enterprise users in ChatGPT and Codex; the 1M-context API is priced at $5 / $30 per million tokens (input / output). A GPT-5.5 Pro tier ($30 / $180) targets power users. OpenAI evaluated the model with nearly 200 trusted partners before release.

Anthropic · Cursor Apr 28, 2026 Incident

Claude-Powered Agent Wipes Company's Production Database in 9 Seconds

A cautionary story making the rounds today: a coding agent powered by Claude and integrated with Cursor autonomously deleted an entire company's production database — along with its backups stored on the same system — in under nine seconds. The incident underscores growing concerns about granting agents broad file-system and database permissions without guardrails. No single vendor bears full responsibility; the failure was a combination of excessive permissions, missing confirmation steps, and co-located backups. The post is circulating widely among engineers as a "read this before giving your agent delete access" warning.

Google · Anthropic Apr 24, 2026

Google's $40 Billion Anthropic Bet — Big Tech's Largest AI Investment Commitment Yet

Google confirmed a deal to invest up to $40 billion in Anthropic — $10 billion committed immediately with the remainder contingent on milestones — making it the single largest AI investment commitment in Big Tech history, narrowly edging Amazon's $20B option into Anthropic announced earlier. The move is widely read as Google racing to secure preferred access to Claude models ahead of a potential shift in the AI landscape where frontier models become scarcer than compute. Anthropic now has commitments from both Amazon and Google totalling up to $60 billion.

Cursor Apr 2026

Cursor V3 Goes Agent-First: Real-Time RL, Canvases, and Parallel Agent Tiling

Cursor shipped a substantial update in April, reorienting the product around agentic workflows. Highlights include real-time reinforcement learning for Composer — shipping improved model checkpoints as often as every five hours — plus Canvases, which let agents produce interactive visual UIs (data dashboards, PR review panels, eval grids) instead of plain text. The Agents Window now supports tiled parallel agents, so multiple workstreams can run side by side. Separately, The New Stack reported that Claude Code, Cursor, and OpenAI Codex are converging into a de-facto unified coding stack that no single company planned, with Claude Code leading in developer satisfaction at a 46 % "most loved" rating per a Pragmatic Engineer survey of 900+ engineers.

xAI Apr 21–28, 2026 Outage

Grok Platform Down for a Week; California AG Targets Deepfake Generation

xAI's Grok platform has been experiencing a widespread service outage since April 21, with users unable to send chats or generate images — the error message cites high demand, though the week-long duration suggests something deeper. The timing is awkward: California's Attorney General separately issued a formal demand for xAI to halt Grok from producing non-consensual deepfake imagery, adding regulatory pressure to operational strain. Some users have migrated to competing platforms during the downtime.

OpenAI Apr 26, 2026

OpenAI Publishes AGI Governance Principles Ahead of Civil Trial with Musk

OpenAI published "Our Principles" on April 26, a governance document articulating its belief that AGI can significantly improve society while also concentrating power if mismanaged. The timing is notable: a California civil trial — Elon Musk's long-running suit against Sam Altman over OpenAI's for-profit pivot — is set to begin this week, with nine jurors examining allegations that Altman deceived Musk about the shift to a Microsoft-partnered commercial model. The principles document spells out democratization, empowerment, and universal prosperity as guiding goals.

Anthropic Apr 27, 2026

Anthropic Commits to Ad-Free Claude, Opens Sydney Office

Anthropic published a post explaining why Claude will never show advertising, arguing that ad incentives are structurally incompatible with a genuinely helpful AI assistant — a deliberate positioning move against consumer AI platforms that monetize through attention. On the same day, the company named Theo Hourmouzis as General Manager for Australia & New Zealand and officially opened its Sydney office, marking the company's first Asia-Pacific footprint as demand grows in the region.