Between October 20 and 22, 2026, defenders faced a stack of disclosures that security writer Johana Bhuiyan dubbed "Black October" in The Guardian: unauthenticated remote code execution in PHP, a supply-chain compromise of LiteLLM, a maximum-severity GitLab flaw, a Brotli out-of-bounds write, and weak intermediate CA keys at eMudhra that prompted the company to leave the global SSL/TLS certificate business.
The PHP flaws, CVE-2026-109618 and CVE-2026-109619, shipped in emergency updates on October 21. Wiz researchers demonstrated remote code execution against WordPress, Drupal, and Joomla installations, plus disclosure of sensitive process memory. The PHP project patched versions 8.2.35, 8.3.36, 8.4.27, and 8.5.12 the same day, and the NSA issued an advisory urging urgent patching. With PHP running a large share of the web, the exposure window was measured in hours, not days.
This matters right now because the same week showed three different failure modes at once: credentials stolen through a trusted dependency, a language interpreter turned into an attack surface, and a certificate authority's trust collapsing under its own weak randomness. Any one of these would have been a bad week. All three landing together is why security teams are treating October 2026 as a turning point rather than a coincidence.
The supply chain proved the weakest link
On October 21, attackers published a malicious LiteLLM version, 1.107.0, to PyPI and compromised Docker Hub images. The window was short, the packages available from 11:39 to 13:23 UTC and the images from roughly 11:46 to 14:20 UTC, but the malware targeted browser credentials, cloud secrets, developer tokens, and AI service API keys. LiteLLM is an open-source gateway for multiple large language model providers, so a single poisoned release reached anyone who pinned their dependency loosely.
The attackers had obtained developer credentials and manipulated the project's release workflow. LiteLLM advised affected users to rebuild environments and rotate exposed credentials. The lesson is blunt: a dependency you trust becomes an attacker's distribution channel the moment its release pipeline is compromised. Pinning versions and verifying signatures would have blocked this specific attack.
Cloud platforms lost customer secrets
Railway disclosed unauthorized access to internal systems on October 22, with encrypted customer secrets exposed and some decrypted using credentials available to the attacker. The attack window opened October 20 at 13:58 UTC; by October 22 at 03:32 UTC Railway confirmed the attacker no longer had access. The company's preliminary finding was that the attacker targeted AI provider API keys, possibly for unauthorized model distillation.
Zeabur published findings from an October 14-15 intrusion in which attackers used leaked internal credentials to reach shared clusters, dedicated servers, and customer environment variables. Environment variables, API keys, service account tokens, and cloud administrative credentials are sensitive because applications use them to authenticate automatically. An attacker who steals one can authenticate to systems far beyond the original hosting platform.
Certificate trust took a hit
On October 17, a researcher reported that private keys for eight intermediate CA certificates operated by eMudhra were cryptographically weak, generated with insufficient randomness during a key-generation ceremony in April 2025. Google Chrome blocked the affected public keys through CRLSet, and Mozilla distributed revocations to Firefox users through OneCRL, both without requiring a browser restart.
eMudhra stated none of the affected intermediates had issued end-entity certificates and found no evidence of a breach of its infrastructure or customer data. On October 22, eMudhra announced it would withdraw from the global SSL/TLS certificate business following the incident and the resulting loss of trust from major technology companies. The distinction matters: no certificates were misissued, yet the trust damage was enough to end a business line.
GitLab and Brotli rounded out the week
GitLab Inc. released fixes on October 21 for a flaw that, under certain conditions, allows remote code execution on self-managed instances. GitLab has not specified what those conditions are, which leaves defenders guessing about their own exposure.
Also on October 21, security patches addressed an out-of-bounds write in Google's Brotli decompression library, tracked as CVE-2026-111497. The flaw affected the Brotli decoder and software incorporating it, including Node.js, Firefox, and Thunderbird. An unauthenticated remote attacker could corrupt adjacent decoder state memory with a crafted compressed data stream, leading to information disclosure, denial of service, or arbitrary code execution.
AI credentials became the prize
October exposed a blind spot in how organizations think about AI. LiteLLM exists to route traffic to multiple model providers, so its compromise was not just a supply-chain problem but an AI infrastructure problem. Railway's attacker appears to have gone after AI provider keys specifically, and OpenCode Console reportedly exposed stored AI provider API keys for some users. The targets are shifting toward the credentials that unlock model access, because those keys are valuable, often long-lived, and rarely rotated on the same cadence as database passwords.
What Black October reveals
The common thread is not the number of incidents but what they reveal about modern attacks. A compromised dependency steals credentials from an otherwise trusted application. A leaked cloud key exposes systems far beyond the original breach. A vulnerable development platform puts CI/CD pipelines at risk. Long-standing weaknesses in widely deployed software become urgent only when researchers or attackers figure out how to exploit them.
The Guardian's reporting framed the period as a new era of AI-enabled hacking, and the LiteLLM compromise fits that framing: attackers targeted AI provider keys specifically. Whether AI tools are accelerating vulnerability discovery or merely amplifying existing weaknesses, the practical response is unchanged. Patch PHP and Node.js now. Rotate any credentials that touched Railway or Zeabur. Audit dependency pins and rebuild affected environments. Verify Brotli and GitLab patch levels. And treat certificate authority trust as something that can evaporate in a week, because eMudhra showed it can.
See also coverage on the Guardian's reporting on Black October for further detail on the PHP vulnerabilities.
For related coverage on this site, see cybersecurity category.
The week left an open question: whether eMudhra's exit from the certificate business is an isolated casualty or the first of several, as browser vendors tighten scrutiny of CA key generation. The answer will shape how much trust organizations place in the certificate ecosystem through the rest of 2026.