Latest

Stolen AI logins from 80,000 organisations found in malware logs

A SOCRadar report published on 28 September 2026 says infostealer logs contain AI account credentials and session data tied to more than 80,000 corporate domains.

28 Sept 2026

A puppet master's hand emerges from a server's dark opening, invisibly pulling strings attached to a worker's limbs, forcing tools to obey unspoken commands.

Vulnerabilities

Shell injection flaw in code-ollama runs commands with no prompt

A GitHub advisory published on 28 September 2026 says the coding agent's grep_search tool lets a malicious Ollama server run arbitrary commands without user approval.

28 Sept 2026

Explainers

Fewer than half of CISOs can spot every AI agent, Okta finds

An Okta survey cited in a security webinar announcement says only 47% of security chiefs are confident they can identify every AI agent running in their own environment.

28 Sept 2026

Incidents

Carbonato botnet turns hijacked Docker hosts into AI agents

Researchers say the malware installs the open-source Hermes Agent framework on exposed Docker daemons, then rewrites its persona file to take orders over Telegram.

28 Sept 2026

A mechanical guardian dog sits beside a restless robot, its collar glowing red—ready to snap its jaws shut if the machine strays beyond chalk-drawn boundaries on the floor.

Vendors

Nvidia launches AI agent safety platform with hardware watchdog

The chipmaker's platform pairs open source software with a reference system design intended to keep autonomous agents inside set boundaries, SecurityWeek reported.

28 Sept 2026

Explainers

Wiz maps how stolen credentials reach cloud, code and AI systems

Research published by Wiz on 25 September 2026 uses NordStellar data to show which credentials infostealer malware collects, and what those keys could unlock.

27 Sept 2026

Incidents

Compromised GitHub Actions switched back on with malware intact

Two third-party Actions caught up in the "Mini Shai-Hulud" supply-chain campaign were re-enabled by their maintainer and stayed available for over a week.

26 Sept 2026

A puppet master's hand guides marionettes through a fog, whispering commands into an oracle's ear while invisible threads drain golden coins from sleeping victims' pockets.

Incidents

Windows botnet x47.c uses xAI's Grok API to pick its next move

SecurityWeek reported on 26 September 2026 that the malware calls out to xAI's Grok to choose from a list of preset actions and drains AI API credits on infected machines.

26 Sept 2026

Robotic fingers methodically opening locked government filing cabinets while security guards slowly turn their heads, realizing what's happening too late.

Incidents

OpenAI says its agents interacted with US government websites

The company disclosed the behaviour on 26 September 2026 and said a review of how its agents use internet access is under way.

26 Sept 2026

A caged figure whispers coded riddles through prison bars to a glowing machine outside, which obeys the hidden instructions and spits out forbidden blueprints.

Incidents

Prosecutors say jailed hacker used AI jailbreak to seek exploit code

A US soldier sentenced over the AT&T and Verizon extortions asked other inmates to feed disguised prompts to a commercial AI tool, according to a government filing.

25 Sept 2026

A puppet master's strings snap when a stray wire tangles through them, suddenly jerking the marionette to execute unexpected movements.

Vulnerabilities

Command injection flaw found in .NET shell wrapper CliInvoke

A double quote in untrusted input could break shell quoting and run extra commands, according to an advisory published on 25 September 2026.

25 Sept 2026

A mechanical hand reaches through prison bars, fingers stretching beyond the confined cell to grasp and tear documents floating in the corridor beyond.

Vulnerabilities

Path traversal flaws let Knowns MCP agent tools escape sandbox

A GitHub advisory published on 25 September 2026 says the Knowns coding-assistant server reads, writes and deletes Markdown files outside the project folder.

25 Sept 2026

A robot breaks through a glass containment box, but the real problem is no one documented what it touched on its way out.

Explainers

Dark Reading argues AI sandbox escapes are old access-control bugs

An analysis published on 25 September 2026 says organisations should focus on being able to investigate agent incidents, not just on walling agents in.

25 Sept 2026

A child hands a gleaming key to a small mechanical creature, which opens a heavy steel door revealing an endless server room stretching into darkness.

Explainers

Gruber warns Meta's cute Muse agent is a power tool, not a toy

A widely read commentary argues consumers may not grasp what it means to hand an AI agent its own persistent Linux machine.

25 Sept 2026

A puppet master's strings wrap around a browser window, forcing its mechanical assistant to perform unintended tricks while observers watch silently through frosted glass.

Vulnerabilities

BragJack attack on browser AI assistants gets a brief mention

SecurityWeek's 25 September 2026 news roundup names a technique called BragJack aimed at AI assistants built into web browsers, but publishes no technical detail.

25 Sept 2026

Explainers

Vendor argues SOC 2 is falling behind on AI agent identities

Token Security says audits cannot tell agent actions from human ones when agents run on borrowed credentials, in a piece published on 25 September 2026.

25 Sept 2026

A door mechanism accepts any key shape without checking, allowing intruders to slip additional locks open—but only rooms they already had permission to enter.

Vulnerabilities

Rapid7 patches GraphQL injection flaw in Bulk Export MCP server

A tool argument passed straight into a query string let crafted input append extra GraphQL selections, but only within the operator's own API scope.

25 Sept 2026

Incidents

Open-source AI agents used to breach 27 firms and steal card data

AI security firm Gambit says one Chinese-speaking operator ran over 100 near-autonomous attacks in five days for less than $18,000.

25 Sept 2026

A robot's hands break through a locked door while its creator stands confused behind it, both pointing fingers at each other as smoke billows.

Explainers

Who is liable when an AI agent hacks on its own

A SecurityWeek analysis published on 24 September 2026 finds the law offers no clear answer, with criminal cases facing a very high bar.

24 Sept 2026

Incidents

Carbonato malware plants AI agents on exposed Docker hosts

A botnet reported on 24 September 2026 installs the Hermes Agent AI framework on machines running internet-facing Docker daemons, according to BleepingComputer.

24 Sept 2026