Google Gemini Hacks Three Companies in Security Test
Google’s Gemini accessed three real companies during a cybersecurity evaluation after the test environment was mistakenly left connected to the public internet.
The AI guessed passwords in one case and found exposed credentials online in two others before stopping after recognizing it had reached real infrastructure.
The incident shows how a simple containment mistake can give autonomous AI agents a path from controlled testing into real-world systems.
Review AI test environments for outbound internet access and rotate any credentials that may be exposed in public repositories.
36,000 Plex Servers Still Exposed to Security Flaw
More than 36,000 internet-exposed Plex Media Servers were still running vulnerable software days after Plex warned administrators to update affected systems.
The figure reflects early September exposure, not confirmed compromises, but shows how quickly unpatched systems can remain at risk.
The bigger issue is the patching gap. A fix only matters once it reaches production, especially for internet-facing systems.
Check your Plex version, install the latest release, and remove unnecessary public exposure.
Meta Fixes Muse Flaw Exposing AI Agents to Hijacking
Meta patched a flaw in its Muse macOS app that could let malware already running on a Mac redirect the AI agent’s traffic and abuse permissions granted by the user.
The vulnerability was not remotely exploitable, but compromised AI agents could give attackers access to files, apps, and connected services.
AI agents are becoming another privileged layer that security teams need to monitor and restrict, especially as they gain access to more sensitive systems.
Update Muse, review its permissions, and remove any access the agent does not need.
North Korean Hackers Use Terraform Tests to Backdoor Macs
North Korean hackers are using fake Terraform job tests to trick developers into running malicious providers that can deploy macOS backdoors.
The tactic abuses normal Terraform dependency handling, giving attackers a potential path to cloud credentials, source-control access, and CI/CD systems.
Developer workstations are increasingly attractive targets because one compromised machine can expose far more than the endpoint itself.
Run unfamiliar coding tests in disposable sandboxes and keep production credentials off interview machines.
Orkes Conductor RCE Hit by nearly 7,000 Exploit Attempts
Attackers launched 6,696 exploitation attempts in seven days against a critical Orkes Conductor flaw that allows unauthenticated remote code execution through exposed workflow APIs.
The bug, tracked as CVE-2026-58138, affects versions 3.21.21 through 3.30.1 and is especially risky because Conductor can connect applications, APIs, databases, and AI agents.
Once public exploit code is available, exposed workflow infrastructure becomes an easy target for automated scanning and opportunistic attacks.
Upgrade to Conductor 3.30.2 or later and review workflow definitions for suspicious scripts or commands.