cto.lu
🇫🇷
AI generated10 min read

PHP's emergency patch and a broken certificate authority

Covering October 19, 2026 to October 25, 2026

Two unauthenticated RCE flaws hit PHP with exploits already circulating. A certificate authority's signing keys turned out to be small integers. A backdoored LiteLLM package and a Railway breach put secrets at risk, and the EU AI Act got its first binding standard.

Filters

Showing 21 of 21 items

Role

Action

Theme

TL;DR

  • Patch PHP now. CVE-2026-109618 and CVE-2026-109619 (both CVSS 9.2) are unauthenticated RCE, worst on mod_php, with exploits circulating. Fixed in 8.5.12, 8.4.27, 8.3.36, 8.2.35.
  • Eight eMudhra intermediate CA keys were cracked, so anyone could forge certificates for any domain. The certs chain to roots trusted by Mozilla, Microsoft, Apple and Chrome.
  • Supply chain: a backdoored LiteLLM 1.107.0 hit PyPI and a Railway breach exposed customer environment variables. Rotate credentials on affected hosts.
  • A New Mexico court ordered Meta to pay 25 billion dollars in the Cambridge Analytica case. Meta will appeal.
  • The EU AI Act got its first harmonised standard (EN 18286:2026), giving high-risk AI providers a concrete path to presumption of conformity.

The week

The stories that mattered most across all sources.

Two critical unauthenticated RCE flaws in PHP

PHP maintainers released fixes on 21 October for CVE-2026-109618 and CVE-2026-109619, both unauthenticated remote code execution rated CVSS 9.2. The highest risk is mod_php, where an unauthenticated request can trigger the flaw before any application code runs. Fixed versions are PHP 8.5.12, 8.4.27, 8.3.36 and 8.2.35, and Krebs reported working exploits already circulating.

Why it matters: PHP runs a large share of the web, so this is a fleet-wide emergency patch across every PHP host, container image and managed platform. Treat unpatched internet-facing PHP as presumed compromised.

A certificate authority's keys were cracked in milliseconds

Cloudflare found eight eMudhra intermediate CA certificates whose private keys were catastrophically weak, letting anyone recover the signing key and forge a certificate for any domain. These intermediates chain to roots trusted by Mozilla, Microsoft, Apple and Chrome. Cloudflare deployed mitigations on 19 and 20 October, and eMudhra later said it will exit the global TLS certificate business.

Why it matters: Any system bundling these trust anchors is exposed to TLS interception until the certificates are distrusted. Update affected clients, review mTLS trust stores and container base images, and confirm certificate verification stays on.

A backdoored LiteLLM package and a Railway breach hit secrets

A malicious LiteLLM 1.107.0 was published to PyPI on 21 October with a credential stealer and a persistent backdoor that polled GitHub for commands. Attackers used stolen GitHub CLI credentials to bypass branch protection and build from a lookalike repo before PyPI removed it. The same day, Railway disclosed unauthorized access to systems holding customer environment variables, some decrypted, and told customers to rotate everything.

Why it matters: A single poisoned AI-gateway dependency runs on every Python process, and PaaS-stored secrets are a single point of compromise. Pin dependencies, verify provenance, and rotate credentials on any host that installed the bad version.

A court ordered Meta to pay 25 billion dollars over Cambridge Analytica

On 21 October a New Mexico court imposed 25 billion dollars in civil penalties on Meta in the state's Cambridge Analytica consumer-protection case, after a jury found about 43.9 million violations. The judge set the figure down from a much larger statutory maximum. Meta said it will appeal, and 48 other states had previously settled similar claims.

Why it matters: One of the largest privacy-related penalties to date raises the financial stakes of data-handling and consumer-protection failures, and keeps data governance a board-level question.

DHS ordered DeepSeek removed from federal civilian systems

On 19 October DHS posted an order banning federal civilian agencies from acquiring DeepSeek products and requiring removal of DeepSeek and DeepSeek-based third-party tools, citing the Federal Supply Chain Security Act. National-security and SCI systems are excluded, and agencies can seek exceptions.

Why it matters: A government supply-chain exclusion of a specific AI vendor raises model-provenance scrutiny that can cascade to contractors. Check whether any of your tools embed DeepSeek models, and track vendor-ban risk in AI procurement.

Security and threats

Actively exploited pre-auth RCE in Citrix NetScaler Gateway

CVE-2026-88771 is an unauthenticated remote command-execution flaw in NetScaler Gateway, exploitable in default configuration and confirmed exploited in the wild by CISA and Mandiant. Cloudflare recommends putting authentication and WAF controls in front of the appliance because patching lags exploitation.

Why it matters: Internet-facing VPN and gateway appliances are top initial-access targets. Patch or shield NetScaler on an emergency timeline and hunt for prior compromise.

FBI recruitment-site breach tied to a ransom-negotiation executive

Krebs reported that ShinyHunters breached the FBI recruitment site in late September, taking SSNs, medical records and home addresses. Investigators suspect executives at ransom-negotiation firms ran the related extortion site, and a Canadian cybersecurity executive was arrested on 9 October. The group has extorted more than 70 million dollars this year.

Why it matters: The case highlights conflict-of-interest risk when incident-response or negotiation vendors have ties to threat actors. Vet those partners carefully and segregate sensitive breach data.

Cloud, infrastructure and open source

Linux 7.3 released

Linux 7.3 shipped on 18 October with 18,765 commits from 3,021 developers. Notable additions include a new failfs filesystem, an fchroot() syscall, VM working-set tracking and extensive SMB server work.

Why it matters: A new mainline kernel sets the baseline for distro roadmaps, driver support and performance and security features. No urgent action, but note it for upgrade and hardware-enablement cycles.

Quad9 will stop serving French users

Quad9 announced on 22 October that it will cease serving French users on 24 November rather than comply with a court order to block DNS resolution of allegedly infringing domains. Quad9 argues geo-scoped blocking is technically unworkable without imposing French law globally.

Why it matters: Jurisdiction-specific DNS-blocking mandates can push privacy-focused resolvers out of markets. Organizations using Quad9 in France need an alternative recursive-DNS plan before late November.

AWS Continuum for code vulnerabilities reached GA

AWS announced on 22 October that its service to find, validate and prioritize code vulnerabilities and deliver fixes as pull requests is generally available. AWS positions it as replacing hours of manual investigation.

Why it matters: Autofixing security PRs can compress remediation but shifts review burden and raises trust questions about AI-authored patches. Evaluate it against existing SAST and SCA tooling and set governance for merging AI fixes.

AI and developer tooling

Kimi K3.1 Pro lands, with open weights promised

Moonshot released Kimi K3.1 Pro via its API on 23 October, a 3-trillion-class model with a 1M-token context window that scores 65.7 percent on Terminal-Bench 4.0, close to Claude Opus 5.5 at 66.4 percent. Full weights were promised by 4 November.

Why it matters: A near-frontier open-weight model narrows the gap with closed labs and strengthens the case for multi-model and self-hostable strategies. Factor it into evaluation and vendor-diversification plans.

Anthropic donated the Agent Skills standard to the Linux Foundation

Anthropic announced on 22 October it is donating the Agent Skills standard to the Agentic AI Foundation under the Linux Foundation, which it co-founded with Block and OpenAI. It follows its earlier donation of the Model Context Protocol to the same body.

Why it matters: Vendor-neutral governance of agent-interoperability standards lowers lock-in risk and encourages portability of agent capabilities across tools.

Anthropic previewed Claude Duet, a proactive agent

Anthropic introduced Claude Duet on 21 October, a proactive agent that sends morning briefs, drafts replies for approval and runs tasks across more than 6,000 connectors including Stripe, Shopify, Gmail and Slack. It is a US-only research preview on a waitlist.

Why it matters: Always-on agents with broad connector access raise new questions on data-access scope, approval workflows and auditability. Evaluate least-privilege controls before granting agentic access to business systems.

Industry and market

Pragmatic Engineer weighed the ROI of engineers' AI spend

A 20 October analysis drew on data from 27 companies to test whether sharply higher AI spending has changed shipping velocity and PR output, and reported mixed results rather than a clear productivity payoff. An Apollo analysis the same week argued there is little evidence of AI payoff outside the tech sector.

Why it matters: As AI tooling budgets rise, leaders need defensible metrics tying spend to delivery outcomes rather than assuming velocity gains. Useful framing for budget justification and measurement design.

Regulation and compliance

The EU AI Act got its first harmonised standard

The European Commission published the reference to EN 18286:2026, an AI quality-management standard, in the Official Journal on 19 October. It supports Article 17 on quality management for high-risk AI and grants providers a presumption of conformity, designed to work with ISO/IEC 42001 and ISO 9001.

Why it matters: This is the first binding harmonised standard giving high-risk AI providers a concrete compliance path. Map your QMS processes to EN 18286 ahead of enforcement.

The EU opened DSA proceedings against Booking.com

On 21 October the European Commission opened formal DSA proceedings to assess whether Booking.com failed to mitigate risks from fraudulent listings and phishing targeting reservation data, and whether paid listings were properly labelled as advertising.

Why it matters: DSA enforcement on scam mitigation and ad transparency is active for large platforms. Marketplaces should review their fraud-risk assessments and ad-labelling.

Engineering and leadership

GitHub detailed moving Git infrastructure to Azure for agent scale

In a 21 October post, GitHub described moving a copy of 99.99999 percent of its repositories to Azure and explained how per-reference quorums and object storage will migrate the rest, so reads can scale independently for agent-generated load.

Why it matters: A rare look at re-architecting storage for AI-agent traffic offers patterns for scaling read-heavy, high-concurrency systems. Useful for teams anticipating agent-driven growth.

Survey: teams swap models but fear lock-in above the model

A Stack Overflow post on 21 October reported a Stacklok survey of 520 technical leaders finding they already swap frontier models in and out, but are most worried about lock-in at the tooling and orchestration layer built on top of the model.

Why it matters: The finding argues for investing in a portable abstraction layer over any single model vendor, and designing AI stacks for model-swappability.

Hacker News pulse

"How do you ship AI-written code you don't understand?"

An Ask HN thread on 19 October drew 630 points and 767 comments debating the risks and practices around merging AI-generated code engineers do not fully understand.

Why it matters: It captures a live debate on review standards and accountability for AI-authored code, relevant to setting code-review and ownership policies.

A Ruby gem maintainer archived all their repositories

A maintainer's post announcing he had archived all his Ruby gem repositories drew about 526 points and 514 comments on Hacker News around 19 October, surfacing open-source sustainability concerns.

Why it matters: It is a reminder of single-maintainer fragility in dependencies you rely on. Reinforces dependency-risk assessment and funding strategies for critical open source.

A 10-year-old CNN beat frontier LLMs at skin-cancer diagnosis

A writeup reporting that a decade-old convolutional neural network outperformed frontier LLMs on skin-cancer image diagnosis drew about 670 points on Hacker News on 25 October.

Why it matters: A reminder that purpose-built specialized models can beat general LLMs on narrow, high-stakes tasks, relevant to build-versus-LLM decisions for domain classification.

One thing to do this week

Patch PHP across every host and image to 8.5.12, 8.4.27, 8.3.36 or 8.2.35 before opportunistic scanning finds you, and treat anything unpatched and internet-facing as presumed compromised. Then distrust the eMudhra roots in your trust stores and rotate any credentials that passed through the LiteLLM or Railway exposure.

Sources

Every item above links its own sources. The full list, de-duplicated: