Anthropic’s doomsday warning to automation push as Claude leak blamed on human error

Anthropic’s Claude Code creator Boris Cherny confirmed that a missed manual step in the deployment process caused the accidental release of 512,000 lines of internal source code. Nobody was fired. His fix: more automation, with Claude checking its own deploys. The irony isn’t subtle—Cherny is the same person who predicted the software engineer job title would start disappearing in 2026, and coding was already “practically solved.”

Anthropic’s Claude Code creator Boris Cherny says the solution to last week’s embarrassing source code leak is—more automation. The same person who predicted the “software engineer” job title would disappear in 2026, and that coding is “practically solved,” is now arguing that a manual deploy step caused the mess, and that leaning harder on AI is the fix.The accidental release compromised roughly 1,900 files and 512,000 lines of Claude Code’s internal source code, touching off a 30-million-view X post, thousands of GitHub forks, and a frantic copyright takedown campaign that itself went sideways—briefly nuking unrelated repos.

One missed step in a manual process did all of this

Cherny attributed the incident to a breakdown in the release process rather than a security breach, explaining that Anthropic’s deployment process has a few manual steps—and the team didn’t complete one of them correctly.“It was a human error,” Cherny wrote on X. “Mistakes happen. As a team, the important thing is to recognize it’s never an individual’s fault—it’s the process, the culture, or the infrastructure.” He confirmed nobody was fired.

What spilled out of those Claude Code’s 512,000 lines of code dump

The code wasn’t just a dry architecture dump. Developers who dug through it found references to unreleased model names—codenames like “Capybara” and “Tengu,” along with model strings pointing to Opus 4.7 and Sonnet 4.8.Some spotted a Tamagotchi-like pet that sits beside the input box and reacts to a user’s coding, along with a “KAIROS” feature that could enable an always-on background agent. Cherny said Anthropic is still on the fence about shipping KAIROS. There was also an internal profanity-tracking dashboard—Cherny confirmed it’s a real UX signal, and that the team calls it the “fucks” chart.

Claude Code creator Boris Cherny’s fix: less human, more machine

His proposed fix: go faster, not slower. More automation, with Claude itself checking the results. That’s a pointed response from someone who, just weeks earlier, told Y Combinator’s Lightcone podcast that coding is “practically solved” and that the software engineer title would start to “go away” this year.Cherny’s remedy cuts against the instinct to add more process after an incident like this. His argument is the opposite—automate the steps that humans keep fumbling, and have Claude check the results. That’s a notable position from someone who, in February, told Y Combinator that coding is already solved and the engineer title is next to go.This is Anthropic’s second security slip in days, after internal files including a draft blog post about an upcoming model were found on a publicly accessible system. The leak, in other words, just made Cherny’s case for him.