Anthropic Claude Code Leak
Something unusual happened at Anthropic and this time, it wasn’t about a new model.
On March 31, a routine update to Claude Code accidentally included a misconfigured debug file. What seemed like a small packaging mistake quickly turned into something much bigger.
Nearly 2,000 internal files and over 500,000 lines of TypeScript code were exposed publicly on npm.
Within hours, a researcher shared the link online. The post exploded — reaching millions — while a GitHub mirror of the code gained tens of thousands of stars almost instantly.
Anthropic later confirmed that no customer data was compromised.
But the damage wasn’t about user data.
It was about exposure.
This incident came just days after leaks related to their unreleased “Claude Mythos” model turning what should have been a normal week into a serious operational setback for one of the industry’s most safety-focused labs.
Why this matters?
This isn’t just a technical mistake.
It highlights how even the most advanced AI companies remain vulnerable to very human errors.
In an industry where companies are building increasingly powerful systems, security and operational discipline are becoming just as important as model capability.
And the irony is hard to ignore — a company known for AI safety briefly lost control of its own code.
Google Releases Gemma 4
While one part of the industry dealt with leaks, Google made a move in the opposite direction opening things up.
On April 2, Google released Gemma 4 under the Apache 2.0 license, marking a major shift from its earlier, more restrictive approach.
This change is bigger than it sounds.
For the first time, developers and companies can freely use, modify, and deploy the model even commercially without legal uncertainty.
But it’s not just about licensing.
Gemma 4 is designed for advanced reasoning and agent-style workflows, and one of its variants is lightweight enough to run on something as small as a Raspberry Pi.
That’s a surprising level of accessibility for a model this capable.
Why this matters?
This signals a major shift in the economics of AI.
Instead of relying entirely on paid APIs and centralized infrastructure, powerful models are becoming freely available and locally deployable.
For developers, this lowers the barrier to building AI products.
For companies, it reduces dependency on large providers.
And for the industry, it intensifies competition especially with models like Meta’s Llama.
AI is no longer just controlled by a few companies.
It’s becoming something anyone can build with.
OpenAI Acquires TBPN
While technology continues to evolve, the battle for influence is becoming just as important.
On April 2, OpenAI announced the acquisition of TBPN, a fast-growing daily tech podcast and live-stream platform.
At first glance, this might seem like a media play.
But the timing suggests something deeper.
As OpenAI moves closer toward a potential IPO, shaping how people understand and talk about AI is becoming strategically important.
Owning a platform that already attracts a large, engaged audience gives OpenAI a direct channel into the tech conversation.
Interestingly, this came on the same day that Anthropic made a very different move acquiring a biotech startup to expand into life sciences.
Two companies.
Two strategies.
Same race.
Why this matters?
This shows that the AI competition is no longer just about building better models.
It’s also about controlling narrative, influence, and positioning.
OpenAI is leaning into media and public perception.
Anthropic is doubling down on deep enterprise integration.
Both are preparing for the same future where AI companies are not just tech providers, but massive platforms shaping entire industries.
The race isn’t just technical anymore.
It’s strategic.
Liked this issue? Get the next one free.
One email on Tuesday. Unsubscribe in one click.