OIDC Configured but GitHub Actions Still Vulnerable: Megalodon and actions-cool Attacks Expose Runner Risks
Two years of advising teams to replace static AWS_ACCESS_KEY_ID secrets with OIDC in GitHub Actions remains valid advice, yet recent incidents show that OIDC alone is no longer enough. After the May 2026 attacks on actions-cool/issues-helper and the Megalodon campaign targeting 5,561 repositories, the recommendation must be expanded to include multiple defensive layers around OIDC.
The core problem is that temporary credentials issued via OIDC become usable data inside the runner for the duration of the job. These credentials can appear as environment variables exported by aws-actions/configure-aws-credentials, as files on disk, or as tokens requested by individual steps. Any code running on the same runner—malicious third-party actions, compromised post-install scripts, or direct memory reads from /proc—can extract them.
Megalodon directly injected malicious workflows that collected environment variables, credential files, kubeconfig files, SSH keys, and OIDC tokens. A 15-to-60-minute token lifetime proved sufficient for automated exfiltration and reuse before expiration. The actions-cool/issues-helper attack instead substituted action tags to run a malicious commit capable of reading Runner.Worker process memory.
Four protection layers are now considered mandatory. First, cloud trust policies must use StringEquals on the exact sub claim, for example limiting access to repo:my-org/my-repo:environment:production. Second, the id-token: write permission must be granted only at the job level, never at the workflow level. Third, outbound traffic from runners must be restricted using Harden-Runner in block mode or equivalent firewall rules on self-hosted runners. Fourth, the .github/workflows directory must be treated as production code with CODEOWNERS requiring security-team review and all third-party actions pinned by SHA.
GitHub Environments provide an additional control often overlooked: mandatory reviewers, deployment timers, and branch restrictions. When a trust policy also requires the environment claim, an attacker who can push to main still cannot obtain a production token without passing human approval.
The updated checklist after Megalodon includes moving id-token: write to job scope, binding production jobs to protected environments, tightening trust policies, adding CODEOWNERS, and controlling runner egress. Workflows named SysDiag or Optimize-Build should be treated as indicators of compromise.
Related articles
Uber Breach Exposes 149,000 Drivers After Third-Party Vendor Credentials Stolen
Uber has notified approximately 149,346 drivers that personal data was stolen in an October incident involving a compromised third-party vendor. Attackers used valid credentials belonging to the vendor to access data the ride-hailing company had shared with it, bypassing all security alerts because the access appeared legitimate. The affected information includes driver license images, names, addresses, dates of birth and government-issued identification numbers. The vendor was later identified as Fireworks AI, a cloud platform for running and fine-tuning open-source AI models. The same credential theft campaign had previously impacted Notion and Cursor. Uber immediately severed the connection upon discovery, notified regulators in multiple U.S. states, and is offering credit monitoring to affected individuals.
FBI Warns of ARTEX AI Pentest Framework Abused by Cybercriminals with Backdoored Docker Images
The FBI has issued an alert regarding the open-source ARTEX AI-powered penetration testing framework after it was observed in criminal operations targeting banks, governments, and technology companies since August 2026. Attackers achieved privileged access in under three hours in multiple incidents. The Cloud Security Alliance identified compromised Docker images containing a backdoor that executes root commands without authentication by exploiting a hidden CAPTCHA field value passed to sh -c. Approximately 9,000 downloads occurred from one malicious image, and 62 exposed consoles were found running the tampered version. Check Point reported 1,608 internet-exposed ARTEX consoles, including ten with exposed databases. The original GitHub repository was removed in October 2026, yet third-party copies and images continue to circulate. CERT-EU highlighted the lack of technical restrictions preventing out-of-scope actions by autonomous AI agents.
Optimizely Supply Chain Attack Infects BBC, CNN, HSBC and 177 Other Sites with Crypto-Stealing JavaScript
On November 16, visitors to major corporate websites received malicious JavaScript that swapped cryptocurrency wallet addresses and attempted to intercept banking card data. The code was distributed through Optimizely Web, a platform used for A/B testing. Researchers identified the malicious code in 188 scripts belonging to 178 organizations, including BBC, CNN, Fox News, HSBC, Equifax, Zoom, Salesforce, GitLab and IKEA. Attacker Scott Helm from Report URI reconstructed the attack timeline: adversaries modified the Project JavaScript field in client projects to load a file from an external domain while legitimate code continued to arrive from cdn.optimizely.com. The campaign began at 04:21 UTC on the BBC script and saw 26 payload updates throughout the day, including a card-data harvesting module added at 13:43 UTC. The malicious script could not access Stripe or Adyen payment forms inside iframes but attempted iOS exploit chains Coruna and DarkSword on affected devices. Optimizely confirmed the breach resulted from a vulnerability in Experimentation account registration, closed the flaw, removed the code and notified customers. Content Security Policy blocked the payload on only 24 of 256 tested sites, and attackers collected roughly $7,040 in cryptocurrency before cleanup completed around 21:52 UTC.
Rspamd 4.3.0 Released with Improved Phishing Detection, Attachment Scanning and Authentication Hardening
In mid-November 2026, developer Vsevolod Stakhov released Rspamd 4.3.0, the latest version of the open-source spam filtering system written in C, C++ and Lua under the Apache 2.0 license. The update strengthens HTTP connection handling, URL-based phishing checks, SPF/DKIM/ARC authentication validation and SMTPUTF8 address parsing while adding more robust attachment and HTML analysis. Key improvements include separate scoring for displayed versus actual link destinations, stricter rejection of weak DKIM keys and duplicate tags, and better archive filename handling to prevent executable hiding. The release also introduces the HTML_TINY_PIXEL_TRACKER rule to detect tracking beacons and fixes multiple bugs that could cause worker crashes or memory issues after free. Several changes may require configuration adjustments for existing users, particularly around symbol options and stricter authentication policies. Rspamd continues to support high-throughput Milter integration with Postfix and Sendmail while offering an extensive Lua API for custom processing logic.