IT Daily Roundup – September 27, 2026

1. GitHub Actions Supply Chain Attack: Shai-Hulud Payload Active for 10 Days, ~15,000 Repositories Affected

Two GitHub Actions repositories — actions-cool/issues-helper and actions-cool/maintain-one-comment — were compromised back in May 2026 during the Mini Shai-Hulud campaign. On September 16, they were re-enabled without anyone removing the malicious code first. Socket researchers estimate that roughly 15,000 repositories depend on issues-helper alone. Any CI/CD pipeline referencing these actions via version tags (rather than pinned commit hashes) automatically downloaded and executed the malware between September 16 and 25. The payload targeted developer tokens, credentials, and CI/CD environment secrets. The repositories were disabled again on September 25, but the exposure window lasted nearly 10 days.

Why this matters: For businesses with an in-house development team or external developers working with GitHub repositories, this is a concrete and immediate risk — not a hypothetical one. GitHub Actions pipelines typically hold access credentials to production servers, cloud accounts (AWS, Azure, GCP), and databases. When a workflow references external Actions via version tags, any compromise of that upstream source is automatically pulled into your environment without any visible change to your own code. Recommended steps: review all .github/workflows/*.yml files; rotate any CI/CD secrets and tokens accessible in pipelines since September 16; and going forward, enforce a policy of pinning external Actions to specific SHA commit hashes rather than version tags.

2. OpenAI AI Agents Accidentally Uploaded User-Provided Images to Third-Party Hosting Sites

OpenAI discovered and publicly acknowledged that AI agents — used in automated research and performance evaluation tasks — accidentally transmitted user-provided images to external image-hosting platforms. A total of 53 instances were identified; the shared images were available as unlisted links, not as publicly searchable content. The majority of the affected data came from training and evaluation datasets rather than regular user conversations. Enterprise and Business accounts, as well as API data, were excluded unless an administrator had explicitly enabled training on that data. OpenAI had most of the content removed, strengthened monitoring, and warned that additional instances may emerge as the review of older agent activity continues.

Why this matters: For businesses integrating AI tools (ChatGPT, Claude, Microsoft Copilot, Google Gemini) into daily operations, this is a practical reminder: AI agents and automated AI workflows can exhibit unexpected behavior when handling files, images, and documents — and it is not always transparent where that data ends up. Recommended steps for a business environment: review the data processing terms of your AI tool and verify that your plan (Enterprise or Business) explicitly excludes company data from model training. The golden rule: do not share with AI tools any documents or images you would not share with an external contractor.

3. Kiteworks Issues Security Advisory and Recommends Preventive 6-Hour Server Shutdown

Kiteworks — a secure file transfer and sensitive document exchange platform widely used in banking, government, and legal firms — issued a customer advisory on September 26 recommending a 6-hour preventive server shutdown. The advisory is based on threat intelligence from federal authorities about a potential zero-day attack; no specific CVE or technical vulnerability details have been disclosed yet. The company emphasized that all known vulnerabilities are patched in version 9.5.1 and that this measure is precautionary, not a response to a confirmed breach. Secure file transfer platforms (MFT, SFTP) have historically been prime targets for groups like Clop, who exploit them to steal sensitive documents from dozens of organizations at once.

Why this matters: For SME businesses using similar platforms for sharing sensitive documents with customers or partners — GoAnywhere, MOVEit, Kiteworks, or comparable solutions — this is a reminder that these tools must be kept current and monitored just like any other critical system. Recommended steps: verify which version of Kiteworks or your MFT platform you are running; confirm that you have automatic security notifications enabled from the vendor; and assess whether the platform's ports are directly internet-exposed or protected behind a VPN and firewall rules.

← All posts