Security advisory: critical PHP and Node.js vulnerabilities are being exploited in the wild. What you need to do →
All months

Radar · 2024-06

June 2024

A PHP-CGI argument injection is exploited within days of disclosure, a third-party polyfill script serves redirects to over 110,000 sites, five plugins are backdoored inside the WordPress.org repository, Chrome moves to distrust Entrust and the June ISM brings large language models into the accreditation baseline.

AI platform and models

Sections
  • notable AI policy

    A national framework for assuring AI in government is agreed — finance.gov.au

    Agreed and released by the Data and Digital Ministers Meeting on 21 June 2024, the framework is a joint Australian, state and territory approach to assuring AI use in government. It is structured as five cornerstones — mechanisms governments should consider — mapped to Australia's AI Ethics Principles, plus practices for applying them.

    Map any AI feature offered to government against the five cornerstones now, so the assurance story exists before it is asked for.

  • fyi Anthropic

    Claude 3.5 Sonnet released — anthropic.com

    Released 21 June 2024 with a 200K-token context window, Claude 3.5 Sonnet was priced at US$3 per million input and US$15 per million output tokens, claiming twice the speed of Claude 3 Opus at the cost of Claude 3 Sonnet and a 64 per cent solve rate on an internal agentic coding evaluation against Opus's 38 per cent. It launched simultaneously on the Anthropic API, Amazon Bedrock and Vertex AI, alongside the Artifacts side-panel workspace.

    Re-baseline any model selection and cost model pinned to Claude 3 Sonnet or Opus.

  • fyi Bedrock

    The newest model reaches Bedrock in Virginia, not Sydney — aws.amazon.com

    AWS announced Claude 3.5 Sonnet availability on Amazon Bedrock on 20 June 2024, listing US East (N. Virginia) as the available region at launch. Asia Pacific (Sydney), which had received Bedrock itself only in April 2024, was not included.

    Confirm which Bedrock region your model identifiers actually resolve to before making residency claims to an agency. The lesson generalises: a frontier model arriving on a platform you are accredited to use does not mean it arrives in the region you are accredited to use it in.

Security and edge

Sections
  • critical PHP

    PHP-CGI argument injection exploited within days of disclosure — nvd.nist.gov

    Published to NVD on 9 June 2024 at CVSS 9.8, CVE-2024-4577 affects Windows installations running Apache with PHP-CGI. Certain code pages trigger best-fit character substitution in the command line passed to Win32 APIs, letting an unauthenticated attacker inject PHP binary options to disclose source or execute arbitrary code. CISA added the CVE to its Known Exploited Vulnerabilities catalogue on 12 June 2024, with a due date of 3 July 2024.

    Confirm no Windows PHP-CGI origin remains below 8.1.29, 8.2.20 or 8.3.8, and add WAF coverage for argument-injection patterns on .php paths as defence in depth. The edge does not save you here unless the pattern is explicitly blocked.

  • notable TLS

    Chrome moves to distrust Entrust TLS certificates — security.googleblog.com

    Announced 27 June 2024, Chrome will stop trusting new TLS server authentication certificates chaining to nine listed Entrust and AffirmTrust roots from Chrome 127, with blocking to begin around 1 November 2024 across Windows, macOS, ChromeOS, Android and Linux. The cutoff keys on each certificate's earliest signed certificate timestamp, so certificates issued before it continue to work. Google cited a pattern of compliance failures, unmet improvement commitments and an absence of measurable progress.

    Inventory certificate issuers across delivered domains. If Entrust or AffirmTrust appears, plan reissuance with a different publicly trusted authority before the existing certificate expires. Certificates issued before the cutoff continue to work until they expire; a replacement issued after it will not.

    Update, September 2024: Google moved the start of blocking to 12 November 2024 to align with Chrome 131. Certificates whose earliest SCT is after 11 November 2024 23:59:59 UTC are no longer trusted.

  • fyi OpenSSL

    OpenSSL 3.3.1 fixes a use-after-free and a slow key check — openssl-library.org

    Released 4 June 2024, OpenSSL 3.3.1 fixed CVE-2024-4741, a potential use-after-free reachable after SSL_free_buffers() is called, and CVE-2024-4603, excessive slowness when checking very long DSA keys or parameters. OpenSSL rated the most severe issue Low.

    Fold into routine base-image and origin patching; no emergency change window needed. The only TLS-stack patch this month was Low severity, which is itself the useful answer.

Standards and compliance

Sections
  • notable ISM

    The June ISM adds two large language model controls — cyber.gov.au

    The June 2024 quarterly Information Security Manual introduces seven brand-new controls, ISM-1918 to ISM-1924, all applicable to every classification. Two are specific to large language models. ISM-1923 requires that "the OWASP Top 10 for Large Language Model Applications are mitigated in the development of large language model applications." ISM-1924 goes further: "large language model applications evaluate the sentence perplexity of user prompts to detect and mitigate adversarial suffixes designed to assist in the generation of sensitive or harmful content."

    If you build or host LLM features for a Commonwealth or state agency, map them against the OWASP LLM Top 10 (the manual cites version 1.1.0) and record the mapping. Then read ISM-1924 carefully — it prescribes a specific detection technique, not a principle, and a system that does no perplexity analysis of prompts does not meet it.

  • fyi OAIC

    OAIC sues Medibank over the security failures behind a breach affecting 9.7 million people — oaic.gov.au

    Filed 5 June 2024, the Australian Information Commissioner's Federal Court proceedings allege Medibank seriously interfered with the privacy of 9.7 million Australians by failing to take reasonable steps to protect personal information between March 2021 and October 2022, in breach of Australian Privacy Principle 11.1. The pleaded chronology: contractor credentials were stolen around 7 August 2022 and used around 23 August to reach Medibank's VPN, which did not require multi-factor authentication; security alerts on 24 and 25 August were not triaged or escalated; and exfiltration began around 25 August. Maximum exposure is up to $2,220,000 per contravention.

    Re-read APP 11.1 obligations against your actual control evidence. The per-contravention framing makes record counts a liability multiplier, which is a direct argument for data minimisation and retention discipline on delivered forms and CMS content.

  • fyi EU AI Act

    The EU AI Act is signed as Regulation 2024/1689 — europa.eu

    Signed by the Presidents of the European Parliament and the Council and dated 13 June 2024, the AI Act was adopted as Regulation (EU) 2024/1689 laying down harmonised rules on artificial intelligence. It enters into force twenty days after publication in the Official Journal.

    Nothing for a domestic-only deployment. Note the date for any vendor with EU market exposure — the risk-tier vocabulary is what Australian AI governance conversations now benchmark against, so it is inherited through procurement even without direct obligation.

CMS and platforms

Sections
  • critical WordPress

    Five plugins backdoored inside the WordPress.org repository — bleepingcomputer.com

    Malicious code was committed to five plugins hosted on WordPress.org between 21 and 22 June 2024, affecting more than 35,000 installations: Social Warfare 4.4.6.4 to 4.4.7.1, Blaze Widget 2.2.5 to 2.5.2, Wrapper Link Element 1.0.2 to 1.0.3, Contact Form 7 Multi-Step Addon 1.0.4 to 1.0.5, and Simply Show Hooks 1.2.1 to 1.2.2. The code created attacker-controlled administrator accounts, exfiltrated the credentials, and injected search-spam JavaScript into site footers. It went undetected until 24 June.

    Check plugin inventories against the affected version ranges, then audit for unexpected administrator accounts and unexplained footer scripts. The update arrived from wordpress.org over the normal channel, so nothing in a WAF or a CDN would have flagged it — treat "came from the official repository" as insufficient provenance.

  • notable Drupal

    Drupal 10.3 is the last feature release of Drupal 10 — drupal.org

    Released 20 June 2024, Drupal 10.3 is the third and final feature release of Drupal 10. It brought a new experimental Navigation module, stable Workspaces and Single-Directory Components, the Recipes system, and new Configuration Actions, Configuration Checkpoints, Default Content and Access Policy APIs. PHP 8.3 is now recommended. Security coverage for 10.3 runs until the Drupal 11.2.0 release in June 2025. Drupal scheduled Drupal 11 for the week of 29 July 2024 and said later Drupal 10 minor releases would be supported until mid- to late 2026.

    Plan the 10.3 upgrade as the prerequisite for Drupal 11, and audit contributed modules for Drupal 11 readiness while you are on it. Recipes and stable Single-Directory Components change how site templates and design systems get packaged and handed over, which matters for anything delivered to an agency.

  • notable WordPress

    WordPress 6.5.5 fixes two XSS paths and a Windows path traversal — wordpress.org

    Released 24 June 2024, this short-cycle release carried three security fixes and three core bug fixes: a cross-site scripting flaw in the HTML API, a second affecting the Template Part block, and a path traversal issue affecting Windows-hosted sites.

    Patch to 6.5.5. If you crawl WordPress to static, re-crawl after patching — a rendered XSS payload baked into published output will keep being served from the edge long after the origin is fixed.

Supply chain, CI/CD and IaC

Sections
  • critical Third-party scripts

    A polyfill CDN script serves redirects to over 110,000 sites — thehackernews.com

    On 25 June 2024 Sansec reported that cdn.polyfill.io — a third-party JavaScript CDN it found embedded in more than 110,000 sites — was serving malicious code, after the domain and its GitHub account were acquired in February 2024. The injected script redirected mobile visitors to betting and adult sites. It was deliberately evasive: firing only on specific mobile devices at specific hours, suppressing itself when it detected an admin session or analytics tooling, and carrying anti-reverse-engineering protection. The original author's position was that modern browsers no longer need the polyfills at all.

    Grep every delivered site for polyfill.io and remove it outright rather than repointing to a mirror. Then inventory all third-party script origins on published pages, apply Subresource Integrity or self-host where you cannot, and add a Content Security Policy that constrains script origins.

  • notable GitHub

    Artifact Attestations reach general availability — github.blog

    On 25 June 2024 GitHub made Artifact Attestations generally available, letting workflows produce Sigstore-signed build provenance linking an artifact to the exact Actions workflow that produced it. Verification runs through the GitHub CLI, and a Kubernetes admission controller shipped alongside to admit only attested artifacts.

    Add the provenance action to release workflows and make verification a gate in the deploy step rather than a manual check. Landing in the same week as the polyfill compromise, it is the practical answer to the question that incident raises — provenance you can verify at deploy time, without standing up your own signing infrastructure.

  • notable Jenkins

    Jenkins plugin advisories leak secrets via logs and unencrypted storage — jenkins.io

    The advisory of 26 June 2024 covers three plugin flaws: CVE-2024-39458, where secrets can appear in default system-log warning messages; CVE-2024-39459, storing credentials unencrypted (only Base64 encoded) on the controller filesystem; and CVE-2024-39460, printing a Bitbucket OAuth access token in the build log.

    Update the three plugins, then rotate any credentials that may have been written to the system log or build logs or stored unencrypted — patching does not un-leak them.

Languages, runtimes and frameworks

Sections
  • notable PHP

    PHP patches all three branches, including a bypass of an April fix — php.net

    PHP 8.3.8, 8.2.20 and 8.1.29 shipped as security releases on 6 June 2024. The ChangeLog records CVE-2024-4577 (argument injection in PHP-CGI, a bypass of CVE-2012-1823) and CVE-2024-5585 (a bypass of CVE-2024-1874). CVE-2024-1874 is recorded on the 11 April 2024 releases as command injection via an array-ish proc_open() $command parameter. NVD says the April fix did not work when the command name includes trailing spaces, which left proc_open() on Windows open to command injection.

    Move all PHP origins to 8.1.29, 8.2.20 or 8.3.8. The useful lesson is the bypass: patching CVE-2024-1874 in April was not sufficient. If you are on 8.1, plan the move to 8.3 — Drupal 10.3 now recommends it.

  • notable Go

    Go fixes Is methods that mis-classify IPv4-mapped IPv6 addresses — groups.google.com

    Go 1.22.4 and 1.21.11, released 4 June 2024, fix two issues. CVE-2024-24790 in net/netip: the Is methods such as IsPrivate and IsLoopback returned false for IPv4-mapped IPv6 addresses that would be true in their traditional IPv4 forms. CVE-2024-24789 in archive/zip: mishandling of certain invalid zip files meant Go's reading diverged from most other implementations, so a crafted file could yield different contents depending on which implementation reads it.

    Rebuild Go binaries on 1.22.4 or 1.21.11, and specifically audit any allow or deny logic built on netip Is methods. Treat that as a trust-decision bug in proxy and edge-function code: an SSRF filter built on those helpers could be fooled by an IPv4-mapped IPv6 address.

  • fyi ASD

    ASD co-authors memory-safety findings for critical open source — cisa.gov

    Published 26 June 2024, joint guidance from CISA, the FBI, ASD's ACSC and the Canadian Centre for Cyber Security reports on the scale of memory-safety risk across a selected set of critical open source projects.

    Nothing immediate. Read it as a signal about where dependency-language questions in government procurement are heading — ASD co-authorship makes this Australian government guidance, on the same secure-by-design track the ISM has been absorbing.

Static site and web tooling

Sections
  • notable Nuxt

    Nuxt 2 reaches end of life — nuxt.com

    Announced 14 June 2024, Nuxt 2 stopped receiving updates — including security and browser-compatibility fixes — on 30 June 2024. The packages remain on npm. The team flagged a final 2.18.0 release as the baseline and pointed users to Nuxt 3, to a commercial extended support arrangement, or to documenting a risk-acceptance position with stakeholders.

    Inventory Nuxt 2 builds. If migration is not imminent, record the risk acceptance explicitly and pin the final 2.18.0 baseline — an unpatched framework is an ISM and assurance problem, not just technical debt, and unlike a CMS there is no vendor issuing backported core patches by default.

  • notable Astro

    Astro adds type-safe environment variables — astro.build

    Astro 4.10, released 6 June 2024, introduced the experimental astro:env module giving schema-declared, type-checked environment variables with an explicit client and server split. The release also extended rewrites to all HTTP methods and made the Container API able to render components outside an Astro site.

    Migrate to astro:env schemas so client-exposed variables are declared rather than inferred, and re-audit existing PUBLIC_ usage. Forcing an explicit declaration of what reaches the client is the single most effective guard against a secret being baked into a published static bundle.