Ten Things That Happened in AI This Week (Week 14)

We spent Easter week in Cyprus. Seven days of pool time, dangerous amounts of ice cream, and weather that couldn’t make up its mind; one day you’re in shorts, the next you’re freezing by the pool in a hoodie. I still managed to fit in some running and strength work between meals, and somehow came home with a mild sunburn despite the shifting conditions. A nice change of pace.

I also managed about two days of not checking my phone before the headlines pulled me back in. That’s all it took. Two days, and the AI landscape had shifted under my feet again. The first week of April 2026 delivered enough news to fill a semester. Here are ten stories that matter if you’re building with, buying, or advising on AI.

1. Anthropic Accidentally Leaks Claude Code’s Entire Source Code

This one is wild. Anthropic pushed version 2.1.88 of Claude Code to npm with a source map file that pointed to a zip archive containing the full, unobfuscated TypeScript source: nearly 2,000 files, 512,000 lines of code, 44 unreleased feature flags, internal model codenames, and a detailed product roadmap. Security researcher Chaofan Shou spotted it, and within hours the codebase was mirrored across GitHub with tens of thousands of forks. The leak revealed features like KAIROS (an autonomous daemon mode), Buddy (a companion system), and anti-distillation measures. For a company that markets itself as the safety-first lab, this is a bad look.

Making it worse: a simultaneous supply-chain attack on the axios npm package meant anyone who updated Claude Code via npm during a specific window may have pulled in a remote access trojan. Anthropic called it “human error, not a security breach.” The internet wasn’t as generous.

Source: Axios, VentureBeat, The Register

2. Microsoft 365 E7 Hits General Availability May 1

Microsoft’s new top-tier enterprise bundle, Microsoft 365 E7 (“The Frontier Suite”), is confirmed for May 1 at $99 per user per month. It bundles E5, Copilot, Entra Suite, and the new Agent 365 into a single SKU. Agent 365, priced separately at $15 per user, is a control plane for observing, governing, and securing AI agents across an organization. Microsoft says tens of millions of agents have already appeared in the Agent 365 registry during preview.

For anyone advising enterprise customers on Microsoft licensing, this is the biggest change since E5 launched in 2015. The component pricing ($60 + $30 + $12 + $15 = $117) makes the $99 bundle look like a deal, but the real question is whether every user in an organization needs all of it.

Source: Microsoft 365 Blog, CNBC, SAMexpert

3. OpenAI Closes a $122 Billion Round at $852 Billion Valuation

That’s not a typo. OpenAI closed the largest private funding round in history, co-led by SoftBank, Andreessen Horowitz, and anchored by $50 billion from Amazon and $30 billion each from Nvidia and SoftBank. About $3 billion came from individual investors through bank channels, and OpenAI is being added to ARK Invest ETFs to broaden its shareholder base ahead of an expected IPO. The company now claims 900 million weekly active users and over 50 million subscribers. Its ads pilot is already pulling in more than $100 million in annual recurring revenue. This isn’t a fundraise; it’s an IPO rehearsal.

Source: TechCrunch, Bloomberg

4. OpenAI Retires GPT-4o and Reshuffles Leadership

OpenAI completed the retirement of GPT-4o from all plans on April 3, the final step in a phased sunset that also took out GPT-4.1, GPT-4.1 mini, and variants of GPT-5. If you haven’t migrated, you’re already behind. Separately, OpenAI reshuffled its leadership: COO Brad Lightcap moves to special projects, former Slack CEO Denise Dresser takes over commercial, and co-founder Greg Brockman steps in on product while Fidji Simo takes medical leave. A lot of moving pieces for a company that’s also trying to go public.

Source: AI and News, The AI Insider

5. Anthropic Acquires Coefficient Bio for $400 Million

Anthropic bought stealth biotech startup Coefficient Bio in a $400 million stock deal. The startup, founded by Samuel Stanton and Nathan C. Frey (both ex-Genentech Prescient Design), was using AI to plan drug research and manage clinical regulatory strategy. The team of about 10 joins Anthropic’s healthcare and life sciences group. This follows Anthropic’s October launch of Claude for Life Sciences. It’s a bet that the same capabilities that make Claude good at code and analysis can accelerate drug discovery. Whether that pans out is another question, but the direction is clear.

Source: TechCrunch, PYMNTS

6. Anthropic Launches a Political Action Committee

Anthropic filed documents to create AnthroPAC, a political action committee that will make contributions to candidates on both sides of the aisle during the midterms. The PAC is funded by voluntary employee contributions capped at $5,000. This comes on the heels of reports that AI companies have collectively poured $185 million into midterm races, and that Anthropic previously funded a Super PAC called Public First with at least $20 million. Anthropic’s political activities are intensifying alongside its ongoing legal battle with the Department of Defense over the government’s use of Claude. The irony? That dispute actually helped Anthropic’s brand in the secondary market, where it’s now the hottest private-market trade around.

Source: TechCrunch

7. Anthropic Introduces a Multi-Agent Harness for Autonomous Development

Anthropic published an engineering design for a multi-agent harness that supports long-running autonomous application development. The approach splits work among planning, generation, and evaluation agents with context resets and structured handoffs. Engineers report 5 to 15 iterations per run, sometimes lasting up to four hours. This isn’t a paradigm shift, but it’s a concrete, well-documented pattern for anyone building production agentic workflows. If you’re in this space, it’s worth reading the actual design.

Source: Let’s Data Science

8. Microsoft Releases Three In-House AI Models, Signals Independence from OpenAI

Microsoft AI, led by Mustafa Suleyman, released MAI-Transcribe-1 (speech-to-text optimized for noisy environments), MAI-Voice-1 (custom voice generation), and MAI-Image-2 (image generation, now top-three on Arena.ai). All three are available on Microsoft Foundry and already integrating into Copilot and Teams. The message is clear: Microsoft is building its own model stack, not just reselling OpenAI.

Suleyman told Bloomberg the goal is to reach state-of-the-art across text, images, and audio by 2027. Microsoft has invested $13 billion in OpenAI but is clearly hedging. They’re also evaluating models from Meta, xAI, and DeepSeek as Copilot alternatives. The partnership lasts until 2030, but the diversification playbook is already running.

Source: TechCrunch, GeekWire, Bloomberg

9. Copilot Cowork Ships with Anthropic’s Claude Under the Hood

Microsoft’s new Copilot Cowork capability, built in partnership with Anthropic, brings long-running multi-step work into Microsoft 365 Copilot. It uses Claude’s model for reasoning and Anthropic’s agentic harness, but runs in the cloud within a customer’s Microsoft 365 tenant, covered by enterprise data protection and grounded by Work IQ. Claude is now also available across the full Copilot Chat experience, not just in Researcher and Excel.

Jared Spataro put it plainly: every 60 days there’s a new king of the hill, so Microsoft is building a platform that isn’t locked to a single model vendor. That’s a smart play. For those of us advising on Microsoft’s stack, the multi-model approach changes the conversation.

Source: Fortune, Microsoft Blog

10. Microsoft Announces $10 Billion AI Investment in Japan

Microsoft committed $10 billion to Japan between 2026 and 2029 for AI infrastructure, cybersecurity, and developer training. The plan includes a partnership with SoftBank and Sakura Internet to provide GPU compute resources from domestic data centers, and a collaboration with NTT Data, NEC, Fujitsu, and Hitachi to train one million AI professionals by 2030. Sakura Internet’s stock jumped 20% on the announcement. This follows Microsoft’s $5.5 billion commitment to Singapore announced the same week.

Microsoft is locking down regional compute capacity at a pace that’s hard to ignore. For European organizations watching from the sidelines, the question isn’t whether AI infrastructure matters; it’s whether we’re moving fast enough to stay relevant.

Source: CNBC

So what’s the thread connecting all of this? Money is flowing at unprecedented scale, but it’s flowing toward companies that are simultaneously building, acquiring, leaking, lobbying, and restructuring. The market is maturing fast, and the organizations that will get the most value from AI aren’t the ones waiting for things to settle. They’re the ones building the governance, the model abstraction, and the evaluation discipline to ride the wave without drowning in it.

Stay curious. Stay critical. And check your npm dependencies.