In this week’s tech update, Dave highlights a major security breach involving Axios, a widely used JavaScript HTTP client with over 100 million weekly downloads. On March 31st, a sophisticated threat actor compromised the lead maintainer’s account on the npm registry and pushed two malicious versions of Axios. Instead of embedding malware directly in the code, the attackers cleverly introduced a phantom dependency called plain cryptojs, which triggered a post-install script. This script deployed a remote access Trojan (RAT) tailored to the user’s operating system, stealthily harvesting sensitive credentials like AWS keys, GitHub tokens, and database passwords before self-destructing to cover its tracks.
The attack has been attributed by Google’s threat intelligence to a North Korean group known as UNC 1069, who have been targeting other security tools as well. Developers who installed Axios between midnight and 3:15 a.m. UTC on that day are urged to assume their machines are compromised, isolate them, and revoke all stored API keys, SSH keys, and cloud tokens. This incident serves as a stark reminder about the risks of supply chain attacks and the importance of using lock files and disabling post-install scripts to mitigate such threats.
The second major story covers a massive accidental source code leak from Anthropic, the AI company behind Claude. On the same day, due to a packaging error during a release on the npm registry, Anthropic inadvertently published a 60MB source map file containing over half a million lines of proprietary code. This leak exposed the internal workings of their AI engineering, including 1900 source files, and was quickly mirrored and forked thousands of times before the company could respond. Unlike a hack, this was a simple but costly mistake in build configuration.
The leaked code revealed fascinating insights into Anthropic’s engineering practices, such as their use of regular expressions for sentiment detection to save on expensive AI token usage, and advanced UI optimizations inspired by game engines to handle high-speed token streaming. It also exposed internal project codenames and features like “undercover mode,” which helps employees contribute to open source without leaving internal traces. This leak comes at a sensitive time as Anthropic prepares for a high-profile IPO valued at $60 billion, with Claude alone generating billions in revenue.
Overall, these incidents underscore the vulnerabilities inherent in modern software supply chains and the critical need for rigorous security practices. From the deliberate compromise of a trusted package like Axios to the accidental exposure of Anthropic’s crown jewels, even the most trusted tools and companies can face significant risks. Developers and organizations must remain vigilant, adopting best practices such as strict version control, script auditing, and rapid incident response to protect their systems and intellectual property. The tech community will be watching closely as more details and forensic analyses emerge.
