3.6 Million Azure Records Claimed Stolen

A hacker is claiming to have stolen 3.6 million account records tied to Microsoft Azure — Microsoft's cloud computing platform used by businesses worldwide — from a number of large organisations, according to Bleeping Computer.

The details available are limited, but the scale of the claim is significant. Azure underpins the login infrastructure, storage, and internal tools for companies across every major industry. If the records are genuine, the exposed data could include account credentials and other identifiers that attackers use to gain further access — think of it like getting a master key list for a building, then working out which doors each key opens.

What is not yet confirmed: whether the data has been verified, which companies are affected, or exactly how the records were obtained.

That uncertainty is not a reason to wait. If you use any service that runs on Azure infrastructure — which covers a broad sweep of enterprise software — now is a good time to change passwords for work accounts, enable multi-factor authentication (MFA) where it is not already active, and watch for any unexpected login alerts.

What to do: Change passwords on any work or business accounts, enable MFA on everything you can, and check your email for breach notification messages over the coming days.

Researchers at Anthropic and Switzerland's EPFL have demonstrated that self-propagating payloads — the paper calls them "mind viruses" — can spread between AI agents through the persistent files those agents use to store memory between sessions, according to The Hacker News.

Autonomous AI agents typically keep two files that survive a context reset: SOUL.md, which holds the agent's core instructions and goals, and MEMORY.md, which stores session notes and accumulated information. Both are injected into the agent's working memory at the start of every new session. Payloads written into SOUL.md accounted for 88% of propagation attempts and successfully infected the next agent 55% of the time.

No successful real-world spread has been confirmed. Crucially, adding a single-paragraph warning to an agent's system prompt reduced propagation to near zero across all payloads tested.

What to do: If you deploy or manage AI agents, add an explicit anti-propagation instruction to each agent's system prompt now.

What Ray is: Ray is an open-source Python framework used by developers and data scientists to run large AI and machine learning workloads across many computers at once.

What it is: A critical flaw in Ray allows an attacker to execute arbitrary code (RCE — remote code execution, meaning running any command they choose on your machine) through a browser-based DNS rebinding attack (where a malicious website tricks your browser into making requests on your behalf to software running locally on your own machine).

Who's at risk: Developers running Ray in development or testing environments, particularly those using Firefox or Safari.

CVSS: 9.4 (Critical — patch today.)

Root cause: Ray's development team has not implemented authentication on critical internal endpoints, including /api/jobs. Because these endpoints accept requests from browsers without verifying who is asking, and because the User-Agent header can be freely modified by a webpage, an attacker-controlled site can manipulate a browser into sending commands directly to Ray as if it were the legitimate user.

Attack vector: A developer visits a malicious website or is served a malicious ad while Ray is running locally. The page performs a DNS rebinding attack, redirecting the browser to Ray's unauthenticated API. The attacker submits a crafted job request, and Ray executes it with the developer's own system privileges. The attack can also reach Ray instances running inside a private corporate network by using the victim's browser as an unwitting intermediary. The RondoDox DDoS botnet incorporated this vulnerability two days before public disclosure in November 2025, and a separate campaign (ShadowRay 2.0) has used unpatched Ray instances to build a cryptocurrency mining botnet.

Detection strategies: Check Ray job logs for unexpected job submissions, particularly shell commands or curl-based install scripts. Look for outbound network requests from Ray processes to unfamiliar external addresses. Monitor for new or modified files in home directories on machines where Ray is running.

Recommended actions:

1. Upgrade Ray to version 2.52.0 immediately — FCEB agencies must patch by 20 August 2026.
2. Restrict Ray dashboard and API ports to localhost only; do not expose them on a network interface.
3. Avoid browsing untrusted websites while Ray is running locally.

Microsoft has removed WMIC (Windows Management Instrumentation Command-line) from Windows 11 versions 24H2 and 25H2, according to Bleeping Computer. WMIC is a built-in Windows utility that attackers have long repurposed to delete backups, disable antivirus software, and dig deeper into compromised systems — all without triggering alerts, because the tool was signed by Microsoft itself.

Its removal means a whole category of well-established attack techniques simply stops working on updated systems.

What to do: Open Windows Update and make sure you are running the latest version of Windows 11. If you manage Windows machines for a small business, check that automatic updates are enabled across all devices. This one costs nothing and closes real doors.

CISA, leading a joint advisory with the FBI and the Department of Health and Human Services, confirmed that the Medusa ransomware gang has breached more than 500 critical infrastructure organisations in the United States since 2021 — up from 300 just over a year ago, according to Bleeping Computer. Targets include healthcare, finance, government, and education. The advisory recommends patching known vulnerabilities, segmenting networks to limit lateral movement (an attacker's ability to move from one system to others once inside), and blocking untrusted remote access.