The internet is having a very bad cybersecurity week

Within a single day, several unrelated security problems came to light and kept security teams occupied. PHP disclosed two critical remote code execution flaws, Node.js released a security update fixing four high-severity issues, a malicious LiteLLM version showed up in public package and container registries, and an Indian certificate authority was found to have eight intermediate keys that anyone could guess. Every one of these problems affects a different layer of the stack, and seeing them together shows how many points of exposure a single server or container image can have at once.

Two critical PHP vulnerabilities

The PHP project released security updates for every supported branch to fix two vulnerabilities that can allow unauthenticated remote code execution. Both received a CVSS v4.0 score of 9.2. One involves out-of-memory handling that can interrupt PHP while internal request data is only partially initialized. The other lets request-owned internal state survive after its memory has been freed, so a later request handled by the same worker can use it. The NSA nicknamed the pair PHP2Shell. Because the flaws live in PHP itself, any web server running an affected build was exposed regardless of the framework or CMS on top. The NSA advised organizations to update right away or take their systems offline until they could, which is why so many public websites disappeared for a while.

Node.js security release

Node.js published its own security release the same day, covering all of its supported lines. It fixed ten vulnerabilities in total, four of them rated high. One is an out-of-bounds write in the bundled Brotli decoder that can lead to remote code execution when an application decompresses attacker-supplied data. Three others affect hostname handling: a DNS function that silently truncates hostnames containing a NUL byte, and two TLS verification bugs where a certificate can be accepted for an IP address it was never issued for.

The Brotli flaw was reported by OpenAI Codex Security, according to the Node.js advisory. The same release also refreshed several bundled dependencies, including a certificate bundle update related to the eMudhra incident described below. TLS validation bugs deserve attention because they affect the check that confirms a client is talking to the server it intended to reach.

The LiteLLM supply chain compromise

LiteLLM is a popular open source proxy that routes requests to many LLM providers. An attacker published a malicious release to PyPI, Docker Hub and GitHub's container registry without touching the source repository, so reading the code on GitHub would have shown nothing unusual. The malicious files were injected during the build.

The payload had two parts. An information stealer collected cloud credentials, SSH keys, environment variables, CI/CD secrets and the AI service keys stored by LiteLLM, while a remote access trojan checked in with the attacker's servers on a schedule and could run any command it received. Python executed both automatically at startup, so nobody had to call the library for the malware to run.

LiteLLM removed the malicious releases and images and published guidance for anyone who had installed them. The incident is a reminder that popular dependencies sit in front of valuable secrets, and that a package can be trusted for years before a single bad release appears.

Weak keys in eMudhra intermediate certificates

The certificate story drew the most attention from security professionals. eMudhra is a certificate authority that the major browsers trust, and eight of its intermediate certificates were generated with private keys that were trivially small. According to analysis from Aikido, every one of the eight keys was below 256, and one of them was literally the number 4. A properly generated key comes from a range of roughly 10^115 values, so an attacker needed at most 256 guesses.

An intermediate certificate can sign a valid certificate for any domain, which lets whoever holds the key impersonate legitimate websites from a position on the network path. Browser vendors blocked the certificates quickly through their own revocation channels. The lasting exposure sits in command-line tools, language runtimes, container images and IoT devices, which carry their own list of trusted roots and only change when someone ships an update.

Other incidents from the same period

Smaller stories arrived around the same time. TiDB Cloud, a database-as-a-service, disclosed a breach. Attackers started exploiting four already-patched Plex Media Server flaws under the name Plex2Shell. BeyondTrust warned customers about a critical zero-day in its Remote Support product that had been exploited for months before the fix.

None of these incidents are connected, but their back-to-back arrival shows the pressure security teams work under. AI played a part in several of the discoveries, and the same tools can help attackers turn a disclosed flaw into a working exploit, which shortens the time defenders have to respond.