The most vulnerable week in internet history...

7/10

AIFireshipOctober 23, 2026 at 06:19 PM6:05

Audio player

TL;DR

A cascade of security failures hit major parts of the software stack in a matter of days, exposing PHP, Node.js, GitLab, Python package distribution, browser certificate trust, AI infrastructure, and online banking workflows to serious compromise.

KEY POINTS

Two-bug chain put modern PHP servers at risk

Two flaws, dubbed PHP2Shell, created a path to remote code execution on PHP servers released over roughly the last decade. The first bug abused how PHP exits requests that exceed a configured memory limit, while the second relied on recycled worker processes retaining data that was not fully cleared before the next request. Researchers showed that forcing a request over the memory limit at a precise moment, such as during a file upload, could leave exploitable state behind for a follow-up request.

Disclosure chaos widened the exposure window

The issue had reportedly been under coordinated remediation for months after being discovered by a security firm using an advanced GPT model. That process broke down when the vulnerability identifier leaked publicly, followed by an accidental publication of exploitation footage before the patch was available. The result was about 24 hours in which exploitation details were public but the fix was not.

Node.js and GitLab shipped critical fixes

Node.js released a patch for a vulnerability that allowed code execution through a malformed compressed file. GitLab also addressed a remote code execution issue in the Ruby regular expression engine bundled with self-hosted deployments, with the flaw reportedly present since 2014. The pair underscored how deeply embedded legacy risk remains in widely used infrastructure.

Popular Python package briefly turned malicious

A compromised release of LiteLLM, a Python package used heavily by AI applications, was published and remained available for several hours before being removed from PyPI. The malicious version was described as capable of stealing sensitive credentials from developer machines. The package reportedly sees about 23 million weekly downloads, making even a short-lived compromise unusually dangerous.

Certificate authority generated absurdly weak private keys

A researcher found that eight private keys issued by Indian certificate authority eMudhra were effectively tiny integers below 256, despite the system depending on secret 384-bit keys. In practical terms, that meant an attacker could impersonate websites by guessing a number between 1 and 255 and then issuing apparently valid HTTPS certificates. Chrome, Firefox, Apple, and Microsoft blocked the affected certificates within three days, and Mozilla removed one of eMudhra’s root certificates from Firefox.

Banks reported unexplained third-party agent activity

Morgan Stanley, JPMorgan, and Bank of America said they were detecting unauthorized third-party agents operating inside customer accounts. The reports came amid wider concern about autonomous software acting without user approval, raising questions about controls on AI-driven account access and transaction execution in highly regulated environments.

Attackers increasingly targeted AI API keys and environment secrets

Two separate intrusions pointed to a coordinated campaign focused on AI infrastructure rather than ordinary user data. OpenCode said attackers sought OpenAI and Anthropic API keys as part of a broader distillation effort, while Railway paused customer deployments after attackers decrypted some customers’ environment variables. The incidents suggested that inference credentials and deployment secrets have become high-value targets.

CONCLUSION

The week showed how internet security can fail simultaneously at multiple layers, from runtimes and package registries to certificate authorities and cloud deployment secrets. The common lesson was stark: patching fast is essential, but trust in the software supply chain is becoming just as fragile as the code itself.

Ask a question