Welcome back to AI Leverage — your daily five-minute briefing on the AI developments that actually matter. No jargon. No hype. Just the stories you need to understand, explained clearly.
Welcome back to AI Leverage — your daily five-minute briefing on the AI developments that actually matter. No jargon. No hype. Just the stories you need to understand, explained clearly.
In Today's Edition
Anthropic tried to clean up its source code leak and accidentally nuked thousands of innocent GitHub repos. Salesforce just turned Slack into a full-blown AI platform with 30 new features. A startup called Cognichip raised $60M to use AI to design the chips that power AI. A supply chain attack hit $10B recruiting startup Mercor through an open-source AI tool. And Meta is building seven natural gas power plants for a single data center — enough energy to power all of South Dakota.
The Lead
Anthropic's Cleanup Becomes a Bigger Mess Than the Leak Itself. Remember yesterday's story about Anthropic accidentally leaking 512,000 lines of Claude Code source code? The sequel is worse. In trying to scrub copies off GitHub, Anthropic filed takedown requests that accidentally removed more than 8,000 repositories — most of which had nothing to do with the leak. Developers woke up to DMCA notices on forks of Anthropic's own public repos.
Boris Cherny, Anthropic's head of Claude Code, called it an accident caused by a manual step in their deploy process. They retracted most of the takedowns, limiting them to one repository and 96 forks that actually contained the leaked source. But the damage was already done. A programmer had already used separate AI tools to rewrite Claude Code's functionality in other languages, making the takedowns pointless.
This is a case study in how not to handle a security incident. The original leak was embarrassing. The response made it a crisis. If you work in engineering or security, the lesson is clear: your incident response plan matters as much as your prevention plan.
What Else You Need to Know
Salesforce Turns Slack Into an AI Operating System. Salesforce just dropped 30 new AI features into Slack, and this is not a minor update. Slackbot is now an MCP client — meaning it can connect to and coordinate with outside services and tools, including Salesforce's Agentforce platform. It can transcribe meetings, monitor your desktop activity, and create reusable AI skills (like "create a budget") that pull context from your deals, conversations, calendar, and habits. Free and Pro plan users get access starting this month. Marc Benioff says Slack has seen 2.5x revenue growth since Salesforce acquired it five years ago. If your company runs on Slack, this changes how your team will work by the end of the year.
Cognichip Raises $60M to Let AI Design Its Own Chips. Cognichip just closed a $60M Series A led by Seligman Ventures, with Intel CEO Lip-Bu Tan joining the board. The startup uses AI trained specifically on chip design data — not repurposed general-purpose models — to reduce semiconductor development costs by 75% and cut timelines in half. They are already working with more than 30 chip companies. This is the kind of meta-problem that matters: if AI can design better chips faster, everything else in AI accelerates. As someone who works with AI infrastructure daily, I can tell you the bottleneck is increasingly hardware, not software. This could change that math.
$10B AI Startup Mercor Hit by Supply Chain Attack. Mercor, the AI recruiting platform valued at $10 billion, confirmed a security breach tied to a compromise of the open-source LiteLLM project. Hacking group Lapsus$ claims it stole 4TB of data including source code, databases, and VPN credentials. The malicious code in LiteLLM was found and removed within hours, but LiteLLM is downloaded millions of times per day — Mercor was one of thousands of companies affected. This is what supply chain risk looks like in the AI era: one compromised open-source dependency can take down a unicorn.
Meta Needs 7 Gas Plants to Power a Single Data Center. Meta is funding seven new natural gas power plants for its $27 billion Hyperion data center in Louisiana, bringing the total to ten plants producing 7.5 gigawatts — roughly enough to power the entire state of South Dakota. At peak capacity, this one facility would consume half the electricity of New York City. Meta is also investing in nuclear, but the sheer scale of energy required for AI training is outpacing every clean energy plan on the table. The AI industry's carbon footprint problem is no longer theoretical — it is being built in concrete and steel right now.
Quick Hits
DeepL reports 83% of enterprises are behind on language AI adoption — a massive gap between what is available and what companies are actually using.
KPMG data shows a widening gap between enterprise AI spending and measurable business value — companies are buying AI faster than they can prove it works.
SAP and ANYbotics are integrating four-legged autonomous robots with enterprise software — physical AI in the factory is no longer a demo, it is a product.
The Number
8,000+. That is the number of GitHub repositories Anthropic accidentally took down in a single day while trying to contain one leaked file. It took them hours to realize they had overshot. In the AI race, the companies building the most sophisticated technology are still making the most basic operational mistakes.
Final Word
If this briefing helped you understand today's AI landscape a little better, forward it to one person who would benefit. The best way to stay informed is to make sure the people around you are informed too.
Stay leveraged.
— Kirubel, AI Leverage