Earlier this year, the head of Anthropic’s blockbuster Claude Code AI agent Boris Cherny boasted that “pretty much 100 percent” of the entire company’s code is AI-generated.
“For me personally, it has been 100 percent for two plus months now, I don’t even make small edits by hand,” he tweeted at the time.
But the glaring cybersecurity implications of giving an AI agent full access over a computer to carry out complex tasks — something experts have been ringing the alarm bells over for a while now — isn’t coinciding during a period of competence for the company: it confirmed on Tuesday that parts of the internal source code for its Claude Code had leaked, which is extremely bad.
“No sensitive customer data or credentials were involved or exposed,” a spokesperson told CNBC, in an apparent effort to focus on the bright side.
The news comes less than a week after news of Anthropic’s upcoming “Claude Mythos” AI model — which the company claimed poses “unprecedented cybersecurity risks” — leaked to the public.
Unsurprisingly, Anthropic attempted to downplay the latest situation and blame human agents, not AI ones, for the leak.
“This was a release packaging issue caused by human error, not a security breach,” the spokesperson added. “We’re rolling out measures to prevent this from happening again.”
A file the company shared on the coding platform GitHub included a link back to the source code, allowing anybody with an internet connection to download it. How the file ended up there in the end, or whether an AI agent could’ve been involved in the process leading up to the leak, remains unclear.
“Claude code source code has been leaked via a map file in their npm registry!” reads an X post, which was viewed tens of millions of times in less than a day.
Anthropic tried desperately to contain the fallout after exposing the source code. As the Wall Street Journal reports, representatives issued copyright takedown requests for more than 8,000 copies and adaptations of the source code, which contains the AI firm’s underlying instructions on how to direct Claude Code.
Ultimately, whether humans — or AI agents — are to blame for the leak almost feels beside the point as the damage has already been done. The exposed data included plenty of proprietary techniques Anthropic uses to point its tool in the right direction.
According toCybersecurity News, the exposed code covers how the company issues authorizations for making changes to resources, “permission enforcement, multi-agent coordination, and even undisclosed feature pipelines.”
As the WSJ points out, competitors will now have an even easier time reverse engineering Claude Code, potentially allowing them to quickly catch up.
The leak could also give hackers a major leg up in their efforts to identify exploitable software vulnerabilities — or find new ways to arm their own instances of Claude Code for nefarious purposes.
“To most of us, this information is useless,” one Reddit user explained. “To people who work for their competitors, you might be able to use this information to understand the ways that they are trying to do things and potentially try and use that information to your advantage.”
“It’s also an exceptional blunder,” they added. “Very embarrassing.”
The incident couldn’t have come at a worse time. The runaway success of its coding assistant has allowed Anthropic to gain a considerable lead as competitors, such as OpenAI, continue to focus their efforts on similar enterprise pursuits.
Meanwhile, the financial pressure continues to build. A recent round of funding is valuing the Dario Amodei-led firm at $380 billion ahead of its rumored IPO later this year.
More on recent Anthropic leaks: Anthropic Just Leaked Upcoming Model With “Unprecedented Cybersecurity Risks” in the Most Ironic Way Possible
The post The Fact That Anthropic Has Been Boasting About How Much Its Development Now Relies on Claude Makes It Very Interesting That It Just Suffered a Catastrophic Leak of Its Source Code appeared first on Futurism.
