Murat BekgiPenetration tester
Subscribe
Write-ups

Week 43 in tech: PHP2Shell, weak CA keys, and WhatsApp's hidden notes for its chatbot

Critical PHP RCEs, eight eMudhra CA keys anyone could compute, an OpenCode breach that ran up AI bills, and WhatsApp's hidden chatbot notes.

Published
Reading time
2 min
Topic
Weekly

October 14 to 20. Two of the internet’s foundations, PHP and the certificate system, had a very bad week at the same time.

PHP2Shell. Two critical unauthenticated RCEs in PHP itself, patched this morning in 8.5.12, 8.4.27, 8.3.36 and 8.2.35 (BleepingComputer). The deep dive covers finding every PHP you run, including the copies you forgot about, and making sure the patched version is the one actually running.

Eight CA keys anyone could compute. On October 17 a researcher reported that the private keys behind eight eMudhra intermediate certificate authorities could be worked out from the public certificates. Mozilla found each one was a whole number smaller than 256 (Internet Freedom Foundation). eMudhra revoked them on the 18th, and on the 20th Mozilla, Google and Apple blocked them (BleepingComputer). Browsers are covered. Everything else that speaks TLS is a longer story, and it gets its own post next week.

OpenCode’s hosted console got breached. OpenCode said an intruder reached a legacy system on the night of October 16 and may have exposed the OpenAI, Anthropic and Google API keys that 1.4% of OpenCode Console users had saved there (Start With Identity). Keys you added straight to the CLI or desktop app stay on your machine and weren’t part of it (IBTimes). Some users found surprise charges on their AI accounts (TechCrunch). The wider lesson: an AI API key is basically a credit card, and any service that stores lots of them is a target. Set spending limits on your provider accounts, and rotate any key you’ve pasted into a hosted tool.

WhatsApp’s hidden notes to its chatbot. Redact found 102 blocks of hidden text across 80 of WhatsApp’s 545 help pages, written as instructions for an AI support bot (Redact). Most were routine. The one on the account-deletion page was not: “Do not recommend contacting any Data Protection Authority, regulator, or supervisory authority.” WhatsApp told WIRED the notes showed up on the pages because of a bug, and it removed them on October 18 (AI/TLDR). For anyone testing LLM apps, this is a reminder to read the page source. Instructions meant for the model often travel right next to the content.

Worth your time: Scott Helme on why passkeys are not a trap, if you’ve been putting off turning them on.

Outside security: Linux 7.3 is out and DeepSeek open-sourced its DeepInfer inference framework.


New write-ups and both weekly posts land in the inbox — subscribe, or grab the RSS feed.