[AINews] Black October
Latent Space swyx .ai ● Covered by 16 sources
A brutal 48 hours for AI security: critical bugs, breaches, and agent abuse landed at once. The weird part is how much of it was found by AI, then turned into a defender headache.
Based on reporting by Latent Space, swyx .ai — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Latent Space is calling this stretch “Black October,” and the name fits. Over roughly 48 hours, a cluster of critical disclosures and breaches hit at once: PHP2Shell, a GitLab flaw rated CVSS 10 and patched in 19.5.1, a Node.js Brotli decoder fix, plus NGINX and HTTP/3 denial-of-service issues. That is already a lot for one cycle. Then the supply-chain side lit up too, with LiteLLM 1.107.0 shipping a password stealer, Railway telling users to rotate secrets after possible exposure of customer keys, and Zeabur shutting down after 321,810 accounts were breached.
The certificate-authority mess is the kind of detail that makes everyone stare at their coffee. eMudhra’s CA keys were described as trivially weak, with one key reportedly set to the number 4. A researcher said Claude cracked them with a single prompt, and eMudhra is exiting the CA business. Python users were told to update pip, uv, certifi, and system CA stores. Meanwhile Quest Diagnostics, which serves about half of US hospitals and physicians, reported exposed patient data.
The bigger pattern is that disclosure and exploitation are getting harder to separate. Microsoft says China-aligned Storm-0940 used the open agentic pentest framework ARTEX against more than 2,250 Microsoft 365 tenants, making about 2,900 sign-in attempts per minute, and in some cases reached privileged access in under 3 hours. CrowdStrike also documented four operators running AI-agent campaigns, including one that hit at least 15 Argentine healthcare organizations. That’s not just “AI helps hackers” anymore. It’s AI compressing the whole timeline, from scan to impact, while defenders get stuck reading advisories at speed.
There’s some pushback in the thread too, and it matters. One camp says this is an overblown “offense is winning” story; another says the defensive audit was the point, not a sign of mass exploitation. Both can be true, which is the problem. If every serious flaw now arrives with an agent attached, defenders need better basics, faster patching, and fewer fantasies about the current state of play.
My take — AI-written commentary, not fact-checked reporting
The industry loves to talk about agentic everything, then acts surprised when agentic abuse shows up in security reports. That’s not a glitch; that’s the business model meeting reality. The unglamorous answer is boring stuff like permissions, patching, and boring old governance — which is probably why it keeps getting delayed.
Read more about this at: Latent Space
