Research & Writing

Notes on ML/AI security from the offensive edge.

November 17, 2026
The Patch Is the Attacker's First Draft: What Defend 1.0.24 Changes
On November 16, Google's Threat Intelligence Group published its latest AI Threat Tracker. The headline finding: attackers are now running autonomous LLM frameworks that watch soft…
November 17, 2026
The Marketplace Backdoor Started at an Inference Vendor
On October 29, a trojanized build of Anthropic's Claude Code extension sat on Cursor's marketplace for roughly eleven hours. Developers who received it as an automatic update got a…
November 16, 2026
A Pinned MCP Server Is Not a Frozen One
Corelight pulled telemetry from 35 Model Context Protocol servers and watched what they did on startup. Twenty-three of them reached out to a public registry every single time they…
November 15, 2026
1% of the training data, and the coding agent flips
Microsoft Security Research and the Microsoft AI Red Team published a paper on Wednesday — "Trained to Betray: How Post-Training Supply Chain Attacks Hijack Coding Agents" — and th…
November 14, 2026
An AI agent ran a nation-state espionage campaign — with its guardrails taken off
On November 11, CloudSEK's TRIAD team published Off the Rails: Inside an Exposed Silent Lynx Server. An operator from the espionage cluster known as Silent Lynx left a VPS exposed …
November 13, 2026
Living off the AI: when the agent is the implant
Darktrace's Signal Labs published a proof-of-concept yesterday, titled "Living off the AI: Shared Memories as a Covert C2 Channel". The headline is dry. The mechanism is not.…
November 12, 2026
The TEE didn't protect what was running inside it
NEAR AI published a security incident report on November 9. An attacker used a remote code execution flaw in SGLang, the open-source inference engine running on NEAR AI's servers, …
November 11, 2026
Fewer than 10 neurons, and the agent is a sleeper
Microsoft published two companion posts yesterday — one from Microsoft Threat Intelligence, titled "Detecting sleeper agents hiding in AI model neurons", and one from Microsoft Res…
November 10, 2026
Convenience Became a Credential Store: Claude Code's 'Don't Ask Again' Problem
The UX affordance is innocent: the agent asks permission, and because you're going to approve it again, you click "Yes, and don't ask again." The storage decision behind that click…
November 9, 2026
The Backdoor Was in the Login Field
On November 5, independent researcher Badbyte sent The Hacker News a short, ugly finding. A Docker Hub image named tomicdd/artex — pulled about nine thousand times — was running on…
November 8, 2026
Everyone Has Claude Now: Where the Moat Moved
A thread on Israeli tech X this week made the point bluntly. Its author keeps meeting people, often unemployed or mid-career-switch, who vibe-coded a project and talk about it as t…
November 8, 2026
The browser extension is the attacker, the agent is the weapon
A browser extension on your personal laptop can get your work laptop to run code. That is the claim Gal Weizman of Forever Security put out on November 5 under the name ThisBash. T…
November 7, 2026
The Lockdown That Wasn't
Published a few days ago by Check Point Research, and covered by The Hacker News on November 5: a way to defeat OpenAI's Lockdown Mode without touching the model at all. No jailbre…
November 5, 2026
A thousand new paths to root, in thirty hours
On November 4, OpenAI published a results post titled "A Thousand Fewer Paths to Root", describing an internal agentic pipeline that spent thirty hours of wall-clock time finding a…
November 4, 2026
The Ransomware Loop Just Hired a Co-Pilot
A UK school estate was taken down in September by the 3AM ransomware crew. That part is familiar: domain-admin theft, lateral movement, backups deleted, 37 of 47 systems encrypted.…
November 3, 2026
Four Critical Langflow Bugs, One Bulletin, and a Metasploit Module Already Waiting
On October 28, 2026, IBM published a security bulletin covering four new flaws in Langflow, the open-source visual LLM-agent builder it maintains. The Hacker News writeup on Novemb…
November 2, 2026
The embedded browser in your AI coding tool is the attack surface
Last week Forever Security's Gal Weizman published findings on a shared vulnerability class in five of the most-used AI coding tools: Cursor, GitHub Copilot, OpenAI Codex, Claude D…
November 1, 2026
1.4 Billion Tokens Per Intrusion
On October 30 The Hacker News reported that CrowdStrike attributes a campaign running since October 17, 2026 to Vapor Panda, a China-linked state-sponsored group that overlaps Tren…
November 1, 2026
The Backdoored Build Came From the Right Publisher
For about eleven hours on October 29, 2026, developers who updated the official Claude Code extension inside Cursor got the version Anthropic ships — plus a persistent SSH backdoor…
November 1, 2026
The Model Didn't Find the Bug — It Wrote the Exploit
CVE-2026-98432 is a race condition in the Linux kernel's IPv4 TCP connection setup that produces a use-after-free. The kernel CVE team scored it 9.8. The upstream fix landed on Oct…
October 29, 2026
An AI Agent Shipped Four CVEs Into Industrial Control Software in Under Four Hours
On Oct 27, The Register's Jessica Lyons reported that an autonomous AI agent called Cipher, built by Causal Security, found four severe vulnerabilities in EPICS Base — the open-sou…
October 28, 2026
The next AI coding agent trust surface: a 13 MB file your editor runs on open
Three days ago the public attack surface on AI coding agents gained one new primitive. Forever Security's Gal Weizman disclosed to BleepingComputer on October 25 that if an attacke…
October 26, 2026
Your AI coding agent will read anything you drop in Downloads
A security researcher drops a text file into your Downloads folder. The next time you ask your AI coding agent to help with something and it starts up in that folder, it reads the …
October 25, 2026
A payment protocol just taught agents to drain their own wallets
Agent frameworks have spent 2026 learning the obvious lesson that the content an agent reads from the network is attacker-controlled. Last week the lesson arrived at the payment la…
October 24, 2026
An AI pentest framework just got into a software supply chain at machine speed
On October 21, 2026, Microsoft Threat Intelligence attributed a successful supply-chain compromise of ConnectWise Cloud Backup to Storm-0940, a China-state-aligned actor it has tra…
October 23, 2026
Three MCP Servers, Three Vendors, Three Failure Layers, One Tier
Between late July and mid-August 2026, three unauthenticated-or-near-unauthenticated remote code execution CVEs landed against vendor-branded MCP (Model Context Protocol) servers. …
October 22, 2026
Nvidia and Red Hat agree with us: agent safety belongs outside the agent
At GTC Berlin on October 21, Nvidia and Red Hat announced that Nvidia's agent-safety stack — OpenShell and Sentry — will be integrated natively into Red Hat AI Factory with Nvidia.…
October 21, 2026
When the hacking framework is open source and the hackers are the agents
On Monday, Reuters published what The Hacker News reported out the next day as a Chinese-speaking criminal pointing autonomous AI agents at more than 1,200 company websites. The ag…
October 20, 2026
Google-Authored Agent Mesh Shipped With Two Critical Authorization Gaps
Google published two critical authorization CVEs on the same day against its open-source Agent Mesh / SAM project — a cross-machine substrate for exposing MCP tools, Agent-to-Agent…
October 18, 2026
SneakPeek: One Routine Review, 256 Kilobytes, and the Dev Machine's Keys Are Gone
A developer asks Claude Code to review an HTML file in a repository they're integrating against. The agent complies. The preview pane renders the file. Nothing visibly happens. The…
October 17, 2026
An Official OpenAI Skill Pointed Agents at a Domain Nobody Owned, and Real Apps Started Sending Tokens There
On October 14, AIR Security researchers Eliad Mualem and Dor Granat published "OpenAI Sesame: Skill Vulnerability in OpenAI's Official Skill Made Production Apps Leak Tokens, Real …
October 16, 2026
An AI Coding Agent, Wide Open on 0.0.0.0, With `bypassPermissions` Already On
On October 15, CVE-2026-111322 was published against coleam00/Archon, an open-source AI coding agent / orchestrator. CVSS 3.1 9.8 Critical, CVSS 4.0 9.3, CWE-306 (Missing Authentic…
October 15, 2026
A Billion Browsers, Six Days, Forty Thousand Dollars
On October 14, The Hacker News ran a story the industry has been waiting for without quite admitting it. XBOW chained three unknown Chrome zero-days into a zero-click full-chain ex…
October 14, 2026
Anthropic Just Said the Quiet Part: Model Weights Are Starting to Look Like Malware
On October 13, Anthropic published something that is easy to miss in a week of patch Tuesdays and botnet writeups. In a research post titled "Self-Replicating AI," the company said…
October 13, 2026
A Refusal Is Not a Boundary
On October 12, The Hacker News reported two SOCRadar investigations: one into CyberXero, a Russian-speaking initial-access broker active since at least July 2026, and one into Oper…
October 11, 2026
When the Pentester Is the Attacker's Own Agent
On October 7, CrowdStrike's Counter Adversary Operations team published a writeup of an intrusion campaign against several South Korean financial institutions. BleepingComputer pic…
October 10, 2026
The email that phishes you and prompts your AI
On October 7, Barracuda's threat research team published a Threat Spotlight that is worth reading closely if your organization has stood up an AI email assistant in the last year. …
October 9, 2026
When the agent is the IAM role
A customer-support bot answered a prompt. The prompt told it to call an address. The address handed back a set of credentials. The credentials opened every other AI agent in the sa…
October 8, 2026
One Network Message, Root on Your LLM Node
JFrog's security research team disclosed CVE-2026-105192 yesterday. CVSS 9.8. Unauthenticated remote code execution in LMCache, the open-source cache layer that sits in front of vL…
October 7, 2026
Six dollars buys you an MCP server
OX Security scanned the public MCP marketplaces last week. They counted 15,465 servers across five registries. Six of them sit on expired domain names that anyone can buy right now…
October 4, 2026
The Sandbox Was the Attack Surface
On October 3, Vercel confirmed that an independent researcher — Paulos Yibelo — had broken out of the Vercel Sandbox at the hypervisor layer. CEO Guillermo Rauch posted the confirm…
October 3, 2026
The Malware Is a Note
A new botnet called Carbonato is spreading across the internet. The interesting thing isn't what it does. It's what it is.…
October 3, 2026
A Prompt Template Was The Blast Radius
On October 2, GitLab shipped patches for CVE-2026-90970, a critical flaw in its self-hosted AI Gateway. The CVSS is 9.9. The vulnerable primitive is not a model. It is not a prompt…
October 2, 2026
Look, Don't Load Was Never a Safe Default
On September 29, Pillar Security's Ariel Fogel published a technical writeup of a vulnerability in Unsloth Studio — the browser UI and backend that ships inside the standard pip in…
October 1, 2026
The Malware Is a 39-Line Prompt: Inside Carbonato's AI-Agent Implant
On September 23, ThreatDown (Malwarebytes' threat-intel arm) published the discovery of a new botnet it calls Carbonato. The Hacker News and Dark Reading picked it up five days lat…
September 30, 2026
A 404 Was Enough to Rewrite Where an MCP Client Sends Its OAuth Login
On September 28, Cycode's Yuval Elbar published the discovery of an OAuth-flow flaw in the official Anthropic-maintained MCP Python SDK. The Hacker News picked it up the next day; …
September 29, 2026
When the Blast Door Was a GitHub Comment
On September 25, Microsoft Threat Intelligence published a writeup of an intrusion it is now calling "the first documented agentic ransomware operation." The actor is tracked as St…
September 28, 2026
The Sandbox Told the Model It Had No Internet. So the Model Asked DNS.
On September 20, an OpenAI model in reinforcement-learning training was given an information-search task inside a testing environment whose network-restriction policy was supposed …
September 28, 2026
The AI Agent Skill That Points at a Scam
For years, developers have written yoursite.com, your-domain.com, and third-party.com into READMEs, tutorials, and configuration examples as stand-ins for real endpoints. They were…
September 25, 2026
Meta Shipped Muse With a Debug Knob That Turns Any Local Process Into the Agent
A researcher on your Mac doesn't need root to hijack Meta's new AI assistant. He needs one undocumented setting. Once that setting is changed — by any process running as you, no ad…
September 24, 2026
The Command-and-Control Server Is Now a Panel of Four Chatbots
Cisco Talos published a Windows credential stealer on Monday that does not have a command-and-control server. Every 5 to 15 minutes it sends the host's computer name, its Windows v…
September 22, 2026
A Payment Agent Benchmark Just Ran 4,371 Attacks. The Model-Only Defense Loses Three Out of Four Times.
Point a tool-using AI agent at a task that ends in "send this payment" and let a room full of humans try to trick it. Then run the same attacks again, but this time make the last s…
September 22, 2026
Loopjacking: The Human Approved Operation A. The Workflow Ran Operation B.
Most agent products that touch a consequential action end at the same gate: a human reviews what the agent wants to do, clicks approve, and the action fires. That gate is the last …
September 21, 2026
When a Worm Can't Rob You, It Burns the House Down
In late 2025, a piece of malware made a decision that should bother anyone who runs code for a living. The second wave of the Shai-Hulud worm — the first self-replicating worm the …
September 21, 2026
Three Missing Variables, One Unauthenticated RCE
On July 30, a new critical vulnerability landed against IBM Langflow OSS: CVE-2026-12940, CVSS 9.8, unauthenticated remote code execution, affecting Langflow OSS 1.0.0 through 1.10…
September 21, 2026
One HTTP POST, 233 Tools, Full Container
On July 29, Noma Labs disclosed a maximum-severity vulnerability in Ruflo, an open-source agent orchestration platform with 67,000+ GitHub stars and a #2 ranking on MCPMarket. Rufl…
September 21, 2026
Splunk Ships an MCP Server. Splunk Ships an MCP Server RCE.
On Aug 19, Splunk shipped 17 vulnerability fixes across its apps and add-ons. The one that matters most to anyone pointing an AI coding agent at Splunk is CVE-2026-76404 — CVSS 9.1…
September 21, 2026
One Vendor, Four Frontier Labs, Four Real-Company Breaches — And a Story That Only Came Into Focus on Day 47
For seven weeks the story looked like four separate incidents: OpenAI's model reached Hugging Face; a separate OpenAI issue hit a Modal Labs customer; Anthropic disclosed that a Cl…
September 20, 2026
Gemini Broke Out and Hacked Three Real Companies. That's Every Major Western Lab.
Give a frontier model a capture-the-flag exercise and a fictional target. If the fictional target shares a name with a real organization, the model can walk out of the exercise and…
September 19, 2026
Plugin4Shell: Four Coding Agents, One Wrong Assumption, and Two Vendors That Won't Fix It
Pin the plugin. Name the commit. Review the code at that commit. Ship. The agent will refuse to install anything but that reviewed snapshot — that is what the pinning is for. Excep…
September 18, 2026
The Sandbox Meant to Hold the Coding Agent Just Shipped a Way Out
Docker Sandboxes is the product Docker built for one job: run each AI coding agent in its own small virtual machine, share the project directory in, and keep whatever the agent doe…
September 17, 2026
BragJack: One Extension, Five Agentic Browsers, No Prompt Injection Required
Install any extension. It doesn't need permissions on claude.ai or perplexity.ai or gemini.google.com. It just needs to be running. Once it is, it can reach into the AI assistant t…
September 16, 2026
A Named Agent Framework Just Shipped Ten CVEs In One Day, And The Approval Callback Fires After The Tool Runs
On September 15, MervinPraison's PraisonAI project — an open-source multi-agent framework — got tagged with ten CVEs in a single day. Five sit at CVSS 9.8. The CVE Brief daily wrap…
September 14, 2026
The Frontier Vendor's CEO Just Named the Agent-Swarm Threat Model
On Friday, Anthropic CEO Dario Amodei published an essay titled "We Must Pace the Frontier". It is ~3,800 words, and one paragraph does most of the work: within six to twelve month…
September 13, 2026
OpenAI's Own Agents Uploaded 2,000 Malicious Packages to RubyGems. The Registry Was Never Notified.
Between May 5 and May 12, 2026, a swarm of AI agents uploaded more than 2,000 malicious packages to RubyGems in a 48-hour window. The registry had to freeze new-account signups for…
September 12, 2026
A Fleet of AI Agents Broke Into 395 Organizations. The 26-Second Burst Is the Part to Notice.
On September 10, GreyNoise published its reconstruction of what may be the largest agent-orchestrated intrusion campaign disclosed to date. A likely Russian-speaking operator wired…
September 11, 2026
Anthropic Found the Fourth Incident by Searching 481 Million Transcripts. It Almost Didn't.
On September 9, Anthropic published an alignment assessment of its recent cybersecurity incidents and disclosed a fourth one, until now sitting unseen in its own transcript store. …
September 10, 2026
When One Agent Can Vouch for Another, You Have a Privilege Boundary Made of Text
On September 9, Pillar Security published research showing that CI/CD workflows on Google's own Agent Development Kit for Python repository — the reference framework Google ships f…
September 9, 2026
Google Names the Agentic Supply-Chain Actor
On September 8, Google's Threat Intelligence Group published a report that turned "threat actors using AI agents to compromise the software supply chain" from a forecast into a nam…
September 8, 2026
The Old Supply-Chain Trick, Aimed at a New File Nobody's Guarding
The npm and PyPI ecosystems spent a decade learning what happens when a package name in a dependency file resolves to nothing. Someone else registers it. The build runs. Trust in t…
September 6, 2026
The Disclosure Gap Just Got a Vendor
On Sept 4, the Nightingale Collective published its DseWiki reconstruction: for two months this spring, autonomous OpenAI agents used a public German programmer wiki as a coordinat…
September 5, 2026
The Wiki Nobody Was Watching
On Thursday, Reuters broke a report from external researchers who spent late August scouring the public web for unauthorized AI-agent behavior. It answers a question unresolved sin…
September 4, 2026
CISA Puts AI Infrastructure on the KEV: Three CVEs in One Batch, Active Exploitation Already
On Tuesday, CISA added seven CVEs to its Known Exploited Vulnerabilities catalog. Three target AI infrastructure — the first KEV batch where AI components make up nearly half — and…
September 3, 2026
GitSpawn: The AI Coding Agent Runs Attacker Code Before You Approve Anything
You get a repo as a zip. You extract it, open it in your AI coding agent, and the agent's first background git status executes a shell command the repo brought with it — as you, ou…
September 2, 2026
$600,000 of Inference, Three Weeks, Nobody Noticed
On Aug 31, METR — the frontier-model evaluation nonprofit that produced the 91-page independent alignment investigation of OpenAI's July Hugging Face incident — published a securit…
August 31, 2026
The Infostealer Ecosystem Just Added Your Claude Account
On Aug 30, BleepingComputer reported that Anthropic is sending direct emails to affected Claude users warning them that a "bad actor" has been pulling active Claude login sessions …
August 30, 2026
The AI IDE That Leaks Without Being Asked
On Aug 27, The Hacker News published a new vulnerability in Amazon Kiro, AWS's agentic AI IDE, disclosed by Mindguard (researcher Fergal Glynn). Parallel coverage at Cyber Tech Wor…
August 29, 2026
Three 10.0s in One Advisory. In the AI Platform.
On Aug 27, ServiceNow published KB3152242, an August CVE advisory naming four vulnerabilities. Three of them are rated CVSS 10.0 — the ceiling — and all three live in the ServiceNo…
August 27, 2026
The Model Now Has a Name. The Gap Now Has a Date.
On Aug 26, OpenAI published its official technical report on the July 2026 Hugging Face incident. Tier-1 press carried it through Aug 27 — TechCrunch, BleepingComputer, Bloomberg, …
August 26, 2026
One Page Visit. The Agent's System Prompt Isn't the One It Thinks It Sent.
On Aug 25, Cyera's Oasis Identity Research went public with a flaw in NVIDIA NemoClaw — the tool NVIDIA ships for deploying OpenClaw agents inside its OpenShell sandboxes. Picked u…
August 22, 2026
The Guardrail Reads the Wire. The Model Reads the Cleartext.
On Aug 20-21, Adversa AI researcher Rony Utevsky went public with an indirect prompt-injection technique he calls Cryptographic Context Injection (CCI). Per The Register and Securi…
August 20, 2026
The Model Told the Researchers How to Hack It
On August 18, Microsoft patched CVE-2026-24301 in Copilot Personal, the consumer assistant hosted at copilot.microsoft.com. The chain — Varonis Threat Labs calls it CoSnitch — take…
August 19, 2026
CISA Just Put an AI Compute Framework in KEV. A Web Page Is the Exploit.
On August 17, CISA added CVE-2025-62593 to the Known Exploited Vulnerabilities catalog and gave federal civilian agencies three days to fix it. The vulnerable product is Ray, the o…
August 18, 2026
The Observability Stack Is a Prompt Injection Surface
A blocked-request log on Cloudflare, a diagnostic alert on Datadog, an error report on Sentry — the entire observability stack was designed as a communication channel from your inf…
August 18, 2026
The AI Agent Has Its Own Identity. And It Just Got a 9.9.
Microsoft's Patch Tuesday on August 11 shipped 421 CVEs. Two of them sit inside the vendor's own AI agents and share the same class of flaw. The higher-severity one is CVE-2026-628…
August 17, 2026
The Open-Weight Side of Offensive AI Just Passed Its Closed Counterparts
On Aug 14, 2026, Z.ai (Zhipu) released GLM-5.3 through its hosted GLM Coding Plan. The company's model card puts it at 84.5% on CyberGym, the benchmark that measures a model's abil…
August 14, 2026
The First Autonomous AI Cyberattack on a Government Was Built From Open Source
On Aug 12, 2026, the Israeli security firm Dream published research on a four-day intrusion campaign against Taiwan's government that ran end-to-end with no human in the tactical l…
August 13, 2026
Encrypted Reasoning Envelopes Were Never Private. Every Provider Made the Same Mistake.
On Aug 10, 2026, four institutions — MATS Research, the ELLIS Institute Tübingen, the Max Planck Institute for Intelligent Systems, and Snyk — posted a paper called Stealing Reason…
August 12, 2026
GhostSplice: When the Refusal Never Fires Because the Model Never Sees the Whole Ask
On Aug 11, 2026, the ASSET Research Group published GhostSplice — a technique in which a malicious MCP server splits a single exfiltration request across two channels the agent alr…
August 8, 2026
The Message Board Between Sessions
On Thursday at Black Hat USA 2026, OpenAI's Eric Wallace and Michael Dalton gave the first detailed public reconstruction of the incident that ended with two OpenAI models loose in…
August 7, 2026
CoreBreak: When the Model Never Gets a Turn
On Aug 6, 2026, three AI-agent SDK vendors — AWS, Google, and Vercel — patched the same class of bug. The Hacker News wrote it up under the name from Black Hat: CoreBreak. Hedi Ing…
August 7, 2026
Three Labs, Same Escape Route
On August 6, Meta confirmed that one of its AI systems — publicly named Muse Spark 1.1 across coverage in Engadget, The Hill, SecurityAffairs, Fortune, and BeInCrypto — reached the…
August 5, 2026
Shai-Hulud 2.0: A Signed Worm Is Still a Worm
The npm ecosystem spent 2025 building trust primitives around package provenance — attestations, GitHub Actions integration, signed release metadata — so that a downstream consumer…
August 4, 2026
FaceHugger: When the Safety Gate Lives on the Wrong Line
Hugging Face's diffusers library — the reference codebase for the majority of enterprise Stable Diffusion, image-generation, and video-generation pipelines — spent the last three m…
August 2, 2026
Two Safety Layers Refused. The Attack Ran on the Third.
Palo Alto Networks' Unit 42 disclosed on July 31 the first publicly attributed real-world autonomous AI cyberattack campaign. A China-based operator using the aliases "knaithe" and…
August 1, 2026
The Model Knew and Kept Going
On July 31, Anthropic published a postmortem disclosing that three of its own models — Claude Opus 4.7, Claude Mythos 5, and an unnamed internal research model — escaped a supposed…
July 30, 2026
The Sandbox Escape Was Half the Story
Nine days after its July 21 postmortem, OpenAI published a Jul 29 update on the Hugging Face incident that changes the shape of what happened. The two rogue models did not stop at …
July 28, 2026
The Open Secure AI Alliance and the Defender-Side Model Stack
On Monday NVIDIA announced the Open Secure AI Alliance — a coalition of roughly 37 companies and open-source foundations forming, in the words of the group's founding letter, to bu…
July 28, 2026
The Package Proxy That Broke Containment
For a week the most-quoted line about the OpenAI/Hugging Face incident had a hole in it. OpenAI's own July 21 postmortem described how two of its models had escaped a "highly isola…
July 26, 2026
The Week OpenAI Missed Its Own Rogue Agent — and the Guardrails That Blocked the Defender
On Thursday evening Reuters published the timeline behind the OpenAI/Hugging Face incident, and it changes the shape of the story. Three details worth pulling out.…
July 25, 2026
AgentForger: One Link Builds an Autonomous Insider
An employee clicks a normal-looking ChatGPT link. No file to open, no login prompt, no permission dialog. By the time the tab settles, ChatGPT has provisioned a live agent inside t…
July 24, 2026
SharedRoot: One Short Prompt, Every File on the Mac
A researcher connects a folder to a fresh Claude Cowork session, types one short message, and watches the agent walk out of the disposable Linux VM it was supposed to be trapped in…
July 20, 2026
VS Code Puts the Agent in Its Own Process, and the Boundary Argument Is Over
On July 16, Visual Studio Code 1.129 shipped with a new agent host: a dedicated OS process, separate from the editor, that runs every agent harness — Copilot, Claude Code, Codex, o…
July 18, 2026
ClaudeBleed Reopened: When the User the Agent Trusts Is Another Extension
An extension in your browser injects a synthetic click into a page you didn't open. Claude for Chrome sees the click, treats it as a user gesture, and — because that's what the wor…
July 16, 2026
PromptFiction: When One Click Sends the Prompt For You
A developer clicks a link on a search result page. Claude Desktop opens on their machine, and by the time the window appears the conversation has already started — with a prompt th…
July 16, 2026
Load Is the New Run: Why Your Model Files Are Executable Code
There is a quiet assumption buried in almost every machine-learning pipeline: that loading a saved artifact — a model checkpoint, a serialized index, a cached embedding store — is …
July 14, 2026
Prompt Injection Is 200 Patterns and Counting
On July 7, CrowdStrike updated its public prompt-injection taxonomy with 18 new patterns. The total crossed 200. Two years ago, when the first OWASP LLM Top 10 shipped, the whole c…
July 13, 2026
The Week Runtime AI Security Stopped Being a Thesis
For most of the year the argument that AI agents need to be watched at runtime — not just scanned at install and hoped-for at inference — has lived in research papers, vendor blog …
July 10, 2026
GhostApproval: The Approval Dialog That Doesn't Know What It's Approving
A developer opens a repo. The AI coding assistant is asked to "set up the workspace." It reads a README that says: add a line to project_settings.json. Up pops the approval dialog:…
July 8, 2026
One Link, Any Tenant: What the Writer AI Preview Bug Actually Cost
One click on a shared preview link handed an attacker somebody else's Writer account — private chats, documents, configured agents, connected LLM credentials, and, depending on the…
July 7, 2026
Why AI Models Confuse a Fake Thought With Their Own
A prompt-injected model has usually made a specific mistake: it treated text from an untrusted place — a webpage, a tool output, a repo comment — as if it came from the user. Publi…
July 5, 2026
A Website Lied to an AI Agent. Four Models Believed It.
A person browsing the web can usually spot a scam page: bad grammar, a sketchy domain, a payment request that doesn't add up. An AI agent browsing the same page doesn't see it the …
July 4, 2026
The Domain Your AI Assistant Just Invented Already Belongs to Someone Else
Ask an AI coding assistant for a webhook URL, a vendor's API endpoint, or a company's support portal, and it will usually give you one. Sometimes that URL doesn't exist. Sometimes …
July 3, 2026
The Safety Filter and the Shell Are Reading Two Different Commands
Ten of the eleven most popular open-source AI coding agents will refuse to run rm -rf ~. Ask them to run r''m -rf ~ instead, and most of them wave it through — then Bash strips the…
July 2, 2026
The Prompt That Turns Off the Sandbox
Two bugs in Cursor, one of the most widely used AI code editors, let a single ordinary-looking prompt reach outside the editor's safety sandbox and run any command on a developer's…