INSPECTING ARCHIVED INTELLIGENCE (OUTDATED VERSION).
Researchers bypass ChatGPT lockdown to exfiltrate chats via phishing
|
2026-11-05 13:59
HIGH
HIGH
AI-ENABLED ATTACK · AUTONOMOUS
MALWARE & BOTNETS
PHISHING & SOCIAL ENGINEERING
Executive Summary
AI-generated
In September 2026, a vulnerability was discovered in OpenAI's ChatGPT desktop application that could enable silent data theft and phishing through zero-click data exfiltration. The flaw was found by Check Point researchers Eli Smadja, who stated that it allowed attackers to bypass Lockdown Mode protections introduced by the AI company in June 2026 as an optional setting. This vulnerability affected users of ChatGPT desktop application, with no specific number provided for how many were impacted. Attackers could exploit this flaw to exfiltrate data or deliver malicious payloads through crafted prompts and shared chats. The attack works by using nested sandboxed frames to bypass Lockdown Mode's security barrier, allowing attackers to access sensitive information without user consent. Following responsible disclosure, the vulnerability was patched on September 18, 2026.
Technical Mitigations AI-generated
• Software Configuration (ATT&CK mitigation for Phishing): Use anti-spoofing and email authentication mechanisms to filter messages based on validity checks of the sender domain (using SPF) and integrity of messages (using DKIM).
Technical Observables 1 IoCs Login Required
Indicators of Compromise (IoCs) Protected
Raw threat telemetry, malicious IP addresses, file hashes, and direct VirusTotal correlation are restricted to authenticated users.
files 1
AI Podcast (EN) detail_available
detail_listen_ai (EN)
Intelligence Metadata
Actors / Malware / CVEs / Campaigns
Target & Sectors
Global Scope
Incident Timeline
February 2026
Attackers exploited the ChatGPT Classic application's zero-click vulnerability, utilizing a spoofed window to deliver malware without requiring user interaction or authentication.
Click on any entity below to view its context and source!
organisation
Technical Details
Technical Details
On February 2026, OpenAI added the
HTML page preview
feature into ChatGPT code blocks.
organisation
OpenAI
Technical Details
On February 2026, OpenAI added the
HTML page preview
feature into ChatGPT code blocks.
infrastructure
Windows
Malware Delivery Through a Spoofed Application Window
Typically, default browser settings prevent JavaScript from opening pop-up windows.
organisation
Video 2
Video 2 – Malware delivery through a spoofed ChatGPT window.
organisation
Zero-Click Exploitation
Zero-Click Exploitation in ChatGPT Classic
We found that
window.open
in the Electron-based ChatGPT Classic application works even without a user gesture.
organisation
Electron
Zero-Click Exploitation in ChatGPT Classic
We found that
window.open
in the Electron-based ChatGPT Classic application works even without a user gesture.
organisation
Video 4
Video 4 – Credential theft.
June 2026
Lockdown Mode was introduced by the artificial intelligence (AI) company in June 2026 as an optional setting to minimize data exfiltration risks arising from prompt injections.
Click on any entity below to view its context and source!
tactic
Exfiltration
Lockdown Mode was
introduced
by the artificial intelligence (AI) company in June 2026 as an optional setting to
minimize data exfiltration risks
arising from prompt injections.
2026/11/05
Threat actors exploited a zero-click vulnerability called PleaseFix to hijack AI agents embedded in agentic browsers, allowing them to execute malicious instructions and bypass ChatGPT Lockdown Mode's security protections.
Click on any entity below to view its context and source!
infrastructure
Windows
The vulnerability allowed for manipulation of the ChatGPT Classic interface, including displaying popup windows with arbitrary content, opening the door to a wide range of phishing attacks.
However, Check Point found a loophole in the manner popup windows are handled during HTML previews that made it possible to escape the confines of the sandbox and bypass Lockdown Mode.
"The HTML preview used nested sandboxed frames, and while ChatGPT intercepted direct attempts to open new windows, the underlying sandbox permissions still allowed popups and permitted them to escape the sandbox," Smadja explained.
organisation
Silent Data Exfiltration
ChatGPT Lockdown Mode Bypass: Phishing, Malware Delivery, and Silent Data Exfiltration.
organisation
Eli Smadja
"The flaw could enable silent chat data theft, phishing, and malware delivery, including zero-click data exfiltration in the ChatGPT desktop application," Eli Smadja, head of research at Check Point, said in a report shared with The Hacker News.
organisation
Check Point
"The flaw could enable silent chat data theft, phishing, and malware delivery, including zero-click data exfiltration in the ChatGPT desktop application," Eli Smadja, head of research at Check Point, said in a report shared with The Hacker News.
organisation
Check Point Research
Check Point Research discovered a vulnerability that allowed attackers to bypass Lockdown Mode, rendering its core protections ineffective.
organisation
Canvas
When enabled, it restricts live web access to cached content, blocks the loading and display of images from the internet, disables network requests from Canvas-generated code, and turns off Deep Research, Agent Mode, live connectors, and file downloads.
organisation
Deep Research
When enabled, it restricts live web access to cached content, blocks the loading and display of images from the internet, disables network requests from Canvas-generated code, and turns off Deep Research, Agent Mode, live connectors, and file downloads.
organisation
Electron
That said, the underlying issue is not limited to the Electron-based desktop environment, highlighting how weakness outside of the scope of the AI model can undermine AI security controls.
infrastructure
Cursor
A set of vulnerabilities in built-in browsers associated with five AI coding assistants, Cursor, GitHub Copilot, OpenAI Codex, Microsoft Visual Studio Code, and Anthropic Claude Desktop that could be exploited to execute code on a host machine, read and steal files, or spoof web browser address bars.
infrastructure
Visual Studio Code
A set of vulnerabilities in built-in browsers associated with five AI coding assistants, Cursor, GitHub Copilot, OpenAI Codex, Microsoft Visual Studio Code, and Anthropic Claude Desktop that could be exploited to execute code on a host machine, read and steal files, or spoof web browser address bars.
organisation
GitHub Copilot
A set of vulnerabilities in built-in browsers associated with five AI coding assistants, Cursor, GitHub Copilot, OpenAI Codex, Microsoft Visual Studio Code, and Anthropic Claude Desktop that could be exploited to execute code on a host machine, read and steal files, or spoof web browser address bars.
organisation
OpenAI Codex
A set of vulnerabilities in built-in browsers associated with five AI coding assistants, Cursor, GitHub Copilot, OpenAI Codex, Microsoft Visual Studio Code, and Anthropic Claude Desktop that could be exploited to execute code on a host machine, read and steal files, or spoof web browser address bars.
organisation
Microsoft Visual Studio Code
A set of vulnerabilities in built-in browsers associated with five AI coding assistants, Cursor, GitHub Copilot, OpenAI Codex, Microsoft Visual Studio Code, and Anthropic Claude Desktop that could be exploited to execute code on a host machine, read and steal files, or spoof web browser address bars.
organisation
Content
Content opened through this route was no longer subject to the same sandbox restrictions, providing a way to circumvent the protections of Lockdown Mode.
organisation
Smadja
"Lockdown Mode was specifically intended to reduce the possibility of data leaving a conversation through potentially malicious outbound requests, yet a weakness in another application feature provided an alternative path around that security boundary," Smadja said.
organisation
SkillJacking
"
The findings coincide with the discovery of a
number of security shortcomings
in various AI-driven tools and services over the past few months to trick them into spilling secrets or carrying out unauthorized actions -
An attack called
SkillJacking
that hijacks dependencies used by skills (repositories, packages, domains, or cloud apps) to compromise every agent installing them.
organisation
LangChain
A technique called
Repeat-After-Me
that targets LangChain LangGraph, Microsoft Agent Framework, Microsoft AutoGen, Google Agent Development Kit 2.0, AWS Strands Agents, and CrewAI and tricks them into running tool calls by means of a "seemingly benign text-repetition request.
organisation
Microsoft
A technique called
Repeat-After-Me
that targets LangChain LangGraph, Microsoft Agent Framework, Microsoft AutoGen, Google Agent Development Kit 2.0, AWS Strands Agents, and CrewAI and tricks them into running tool calls by means of a "seemingly benign text-repetition request.
organisation
Microsoft AutoGen
A technique called
Repeat-After-Me
that targets LangChain LangGraph, Microsoft Agent Framework, Microsoft AutoGen, Google Agent Development Kit 2.0, AWS Strands Agents, and CrewAI and tricks them into running tool calls by means of a "seemingly benign text-repetition request.
organisation
Google
A technique called
Repeat-After-Me
that targets LangChain LangGraph, Microsoft Agent Framework, Microsoft AutoGen, Google Agent Development Kit 2.0, AWS Strands Agents, and CrewAI and tricks them into running tool calls by means of a "seemingly benign text-repetition request.
organisation
AWS Bedrock AgentCore
"
A vulnerability called
AgentCorruption
that could allow an attacker to interact with an exposed AI agent running on AWS Bedrock AgentCore and leverage prompt injection techniques to influence that agent and extract its
IMDS
credentials, and then use them to access other agents, read private conversations, steal connected-tool credentials, and manipulate agent behavior.
organisation
SalesBleed
A zero-click vulnerability called
SalesBleed
that could be exploited to extract sensitive account data out of Salesforce Agentforce by leveraging the Web-to-Lead form to plant an indirect prompt injection payload.
organisation
Meta Muse's
A proof-of-concept (PoC) attack called
not-a-mused
that allows an unprivileged local attacker (or process) to redirect Meta Muse's dictation traffic and abuse the permissions granted to the AI assistant to perform arbitrary actions.
organisation
Air Security
"The agent checks out the exact commit the marketplace pinned but never verifies it landed there, so an attacker who controls the plugin's repo makes the checkout resolve to malicious code while the pin still looks honored," Air Security said.
organisation
Google Chrome
"
An attack called
BragJack
that can be exploited to hijack the AI agents built into browsers like Google Chrome, Microsoft Edge, Opera Neon, Perplexity Comet, and Anthropic Claude in Chrome by using an already installed malicious browser extension in conjunction with
prompt forcing
.
organisation
Microsoft Edge
"
An attack called
BragJack
that can be exploited to hijack the AI agents built into browsers like Google Chrome, Microsoft Edge, Opera Neon, Perplexity Comet, and Anthropic Claude in Chrome by using an already installed malicious browser extension in conjunction with
prompt forcing
.
organisation
Opera Neon
"
An attack called
BragJack
that can be exploited to hijack the AI agents built into browsers like Google Chrome, Microsoft Edge, Opera Neon, Perplexity Comet, and Anthropic Claude in Chrome by using an already installed malicious browser extension in conjunction with
prompt forcing
.
organisation
Workflow Identity Hijacking
A technique called
Workflow Identity Hijacking
allows an attacker to bypass standard controls by sending normal, benign requests through an unauthenticated entry point, causing the enterprise AI pipeline to process the request and execute the action as designed.
organisation
PleaseFix
"
A zero-click vulnerability class called
PleaseFix
that could be exploited to hijack
AI agents
embedded in
agentic browsers
by placing malicious instructions inside emails, calendar invitations, or web pages, and exploit a technique referred to as intent collision to cause the agent to "act on the attacker's behalf using the user's own identity, permissions and access.
organisation
GitHub
"
A
set of vulnerabilities
impacting
AI coding agents
from Anthropic, Google, and OpenAI that could be exploited to trick them into handing over a victim's secrets, achieving persistent instruction poisoning by modifying the AGENTS.md file, and compromising software releases through prompt injections placed inside untrusted user inputs, like GitHub issues.
Tactical Metrics
Metrics
infrastructure
Windows
Affected Product
Click for context!
The vulnerability allowed for manipulation of the ChatGPT Classic interface, including displaying popup windows with arbitrary content, opening the door to a wide range of phishing attacks.
Malware Delivery Through a Spoofed Application Window
Typically, default browser settings prevent JavaScript from opening pop-up windows.
However, Check Point found a loophole in the manner popup windows are handled during HTML previews that made it possible to escape the confines of the sandbox and bypass Lockdown Mode.
"The HTML preview used nested sandboxed frames, and while ChatGPT intercepted direct attempts to open new windows, the underlying sandbox permissions still allowed popups and permitted them to escape the sandbox," Smadja explained.
Metrics
infrastructure
Cursor
Affected Product
A set of vulnerabilities in built-in browsers associated with five AI coding assistants, Cursor, GitHub Copilot, OpenAI Codex, Microsoft Visual Studio Code, and Anthropic Claude Desktop that could be exploited to execute code on a host machine, read and steal files, or spoof web browser address bars.
Metrics
infrastructure
Visual Studio Code
Affected Product
A set of vulnerabilities in built-in browsers associated with five AI coding assistants, Cursor, GitHub Copilot, OpenAI Codex, Microsoft Visual Studio Code, and Anthropic Claude Desktop that could be exploited to execute code on a host machine, read and steal files, or spoof web browser address bars.
Intelligence Sources
Check Point Research
2026-11-05
Zero Day Fans
2026-11-05
The Hacker News
2026-11-05
Unpublish from Social Media?
Are you sure you want to delete this podcast video from all synchronized social networks (YouTube, Facebook, Threads)?
Important:
Due to Meta API restrictions, Instagram Reels cannot be deleted automatically via API by third-party apps.
View Profile to Delete Manually
View Profile to Delete Manually
Tactical Intelligence
Report Intelligence Issue
Podcast Options
Generate
Reset / Delete
Incident Version History
CURRENT VERSION
Last Updated: 2026-11-06T13:10
Comprehensive Tactical Telemetry
Highly Correlated Entities
34x
organisation
Identified Entity
Silent Data Exfiltration
entity
3x
timeline
Temporal Reference
September 18, 2026
date
3x
tactic
Cyber Operation Type
Phishing
tactic
3x
tactic
MITRE ATT&CK Technique
T1566 - Phishing
technique
3x
infrastructure
Affected Product
Windows
software
Contextual Telemetry
Context Block
2 METRICS
target region
Target Country
United States
country
general metric
Agent Development Kit
2
agent development kit
Click on any entity below to view its context in the main text!
Selective Unpublish
Select the networks you want to remove this post from. The system will try to delete the real post through the API and clean the database so you can publish it again.
By navigating this website, you accept the use of strictly necessary technical cookies for session security and basic platform functionality. We do not use tracking or advertising cookies. Read our Privacy Policy.
Generating report…
0s