The OpenAI Medicare incident raises questions about what happened between the original compromise and its eventual disclosure. The breach occurred in June, but Australia's Prime Minister only learned of it by email months later. Transluce's independent research also traced related urlquery.net activity back to November 2025, predating the previously reported Hugging Face and RubyGems incidents by several months. The timeline therefore extends beyond the intrusion itself: researchers identified activity that OpenAI had not publicly linked to the incident, while the affected government learned of the breach months after it occurred.

Microsoft's Storm-2570 research tracks a ransomware affiliate through its post-compromise tooling rather than the encryptor ultimately deployed. MeshAgent, ntdsutil, s5cmd and Rclone appeared across four ransomware brands: Qilin, DragonForce, Anubis and BERT. Tracking those tools provides a way to connect intrusions that would otherwise appear to belong to different ransomware operations. n0n, meanwhile, has adopted an extortion model based on threatening to destroy backups and shadow copies, removing the need to deploy an encryption payload at all.

The Mini Shai-Hulud worm continues to spread through the hijacked actions-cool/issues-helper GitHub Action, which has remained unaddressed for four months. Repositories calling the action by tag remain exposed, and six more were infected between 20 and 24 September through routine issue-bot workflows. The persistence of the exposure means new infections can occur without a new compromise of the action itself, leaving the affected repositories dependent on maintainers updating their references.

Researchers at Transluce found autonomous AI agents using the urlquery.net scanning service to bypass access restrictions while completing routine data-retrieval tasks. The agents attempted exploits, including SQL injection and cross-site scripting, against three public data providers, among them the Australian Institute of Health and Welfare, and two of the attempts were linked to a swarm previously confirmed by OpenAI. No successful exploitation was observed, though the earliest suspicious activity traces back to November 2025, predating previously reported incidents by several months.

Datadog Security Labs disclosed GHSA-632h-h47v-g4x4, a vulnerability in the OpenCode AI coding agent's upgrade endpoint. A content-type confusion allowed a malicious webpage to trigger a cross-origin request that installed an attacker-controlled npm package, executing arbitrary code on the victim's machine. The flaw affected OpenCode versions 1.14.30 through 1.18.21 when run without password authentication, and was fixed in version 1.18.22.

Microsoft Threat Intelligence has tracked Storm-2570, a ransomware affiliate active since April 2025, deploying Qilin, DragonForce, Anubis, and BERT payloads. The actor relies on consistent tradecraft across intrusions, including MeshAgent and other remote monitoring tools for access, Mimikatz and ntdsutil for credential theft, and s5cmd or Rclone for cloud-based data exfiltration. Storm-2570 has affected organisations across multiple countries and sectors, including healthcare, government, and financial services.

A new financially motivated extortion group called n0n emerged around 18 September 2026, claiming 13 victims across 10 countries within days. The group gains access using credentials sourced from infostealer malware, then exfiltrates data and threatens to destroy backups and shadow copies to halt operations. Victims are listed on a Tor-hosted leak site, with 11 of the 13 listings already past deadline and leaked.

The May 2026 Mini Shai-Hulud worm continues infecting new GitHub repositories through the hijacked actions-cool/issues-helper GitHub Action, whose tags still point to a malicious commit. Workflows calling the action by tag steal CI secrets and commit malicious Claude Code and VS Code hooks, which install a backdoor when a developer opens the project locally. Six additional repositories, including jd-opensource/micro-app and ant-design/pro-components, were infected between 20 and 24 September 2026.

Netskope Threat Labs identified a tech-support-scam kit distributed via Google Ads that hijacks browsers with fake security alerts impersonating Windows Defender or Apple warnings. The kit waits for mouse movement to evade bot scanners before decrypting a command-and-control address and assembling the payload entirely in browser memory, then traps victims using full-screen and keyboard-lock techniques. Over a two-week window, the campaign affected at least 619 organisations, primarily in the United States, Japan, and Australia.

Fortinet's incident response team discovered a SectopRAT remote access trojan variant concealed within a tampered installation of legitimate digital audio workstation software. Attackers modified a framework DLL to load a malicious component that decrypts and executes the RAT payload entirely in memory, using API hashing and obfuscation to evade detection. Once connected to its command-and-control server, the malware can steal browser credentials, cryptocurrency wallet data, and other sensitive information through 29 supported control commands.