Live Feeds
● TRACKER Updated 9d ago · 55 sources tracked

A Flaw in ChatGPT’s Mac App Could Have Let Hackers Grab Sensitive Data

OpenAI has fixed a high-severity security vulnerability in ChatGPT that allowed attackers to bypass Lockdown Mode protections and access private user data. Discovered by Check Point Research, the flaw exploited HTML page previews inside code blocks to execute malicious code, facilitating silent chat theft, phishing, and malware delivery. OpenAI patched the vulnerability before public disclosure, and researchers have confirmed that the exploit is no longer functional. The incident underscores ongoing security risks involving client-side data handling and desktop artificial intelligence applications.

🎙️

Listen to Live Briefing

Real-time synthesized voice briefing · Live Feeds Desk

⏱ ~3 min
Speed:
RSS Source map (55)
⚡ Key Developments & Real-Time Context
Text size:
  • ✓ Check Point Research identified a vulnerability that could bypass OpenAI's ChatGPT Lockdown Mode.
  • ✓ Lockdown Mode was introduced by OpenAI in June 2026 as an optional setting to minimize data exfiltration risks arising from prompt injections.
  • ✓ OpenAI patched the vulnerability following responsible disclosure, and researchers confirmed the demonstrated technique no longer works.
  • ✓ The vulnerability stemmed from HTML page previews inside ChatGPT code blocks that allowed HTML content to be rendered directly within the application.
🛡️ Source Corroboration: 55 independent reporting domains (100% confidence) ⏱ Read time: ~2 min

What changed

Security researchers disclosed technical details of a patched ChatGPT vulnerability that bypassed Lockdown Mode protections introduced in June 2026.

Live updates

  1. OpenAI Patches ChatGPT Lockdown Mode Bypass Vulnerability

    OpenAI has fixed a high-severity security vulnerability in ChatGPT that allowed attackers to bypass Lockdown Mode protections and access private user data. Discovered by Check Point Research, the flaw exploited HTML page previews inside code blocks to execute malicious code, facilitating silent chat theft, phishing, and malware delivery. OpenAI patched the vulnerability before public disclosure, and researchers have confirmed that the exploit is no longer functional. The incident underscores ongoing security risks involving client-side data handling and desktop artificial intelligence applications.

    Why it matters

    OpenAI introduced Lockdown Mode in June 2026 as an optional security setting designed to minimize data exfiltration risks from prompt injections. The newly patched vulnerability allowed attackers to slip past these defenses using a hidden pop-up trick to copy full chat histories and show fake windows. Although sandboxing is used to isolate user-supplied HTML and JavaScript from local storage and cookies, researchers found ways around these measures until OpenAI deployed a fix.

    What is confirmed

    • Check Point Research identified a vulnerability that could bypass OpenAI's ChatGPT Lockdown Mode.
    • Lockdown Mode was introduced by OpenAI in June 2026 as an optional setting to minimize data exfiltration risks arising from prompt injections.
    • OpenAI patched the vulnerability following responsible disclosure, and researchers confirmed the demonstrated technique no longer works.
    • The vulnerability stemmed from HTML page previews inside ChatGPT code blocks that allowed HTML content to be rendered directly within the application.

    What to watch next

    • Further technical disclosures from OpenAI regarding how the Lockdown Mode sandbox was bypassed.
    • Additional security audits of desktop artificial intelligence applications.
    Sources used for this update (4)
    1. thehackernews.com — Researchers Bypass ChatGPT Lockdown Mode to Exfiltrate Chats ...
    2. lapaasvoice.com — OpenAI Fixes ChatGPT Lockdown Mode Bypass Found by Check Point
    3. cybersecuritynews.com — Hackers Could Bypass ChatGPT Lockdown Mode to Steal ...
    4. www.expresscomputer.in — Security flaw could expose ChatGPT conversations to data ...
    confidence 100%
  2. ChatGPT Flaw Exposed Private Conversations and Sensitive Data

    Security researchers discovered a vulnerability in the ChatGPT macOS application that exposed sensitive user data and private conversations through signed-in browsers. The security flaw allowed unauthorized access when users opened malicious shared conversation links, executing attacker-controlled code within authenticated sessions. OpenAI patched the high-severity vulnerability before public disclosure. The incident highlights ongoing security challenges regarding client-side data handling in desktop artificial intelligence applications, demonstrating how malicious links can compromise local privacy if proper input sanitization and zero-trust protocols are absent.

    Why it matters

    Desktop applications that integrate with cloud-based artificial intelligence services present unique security perimeters. When client-side vulnerabilities allow external links to execute code inside authenticated user sessions, the boundary between local protection and remote data storage blurs. Security teams continue to examine how client-side zero-trust sanitization can prevent unauthorized data transmission to external servers.

    What is confirmed

    • Researchers found a ChatGPT flaw that exposed private chats through signed-in browsers.
    • OpenAI fixed the vulnerability before public disclosure.

    Still unconfirmed

    • Client-side zero-trust sanitization is confirmed to be fully deployed across all future versions to completely stop data leaks.

    What to watch next

    • Technical post-mortems detailing the exact mechanism of the macOS app vulnerability
    • Additional updates from OpenAI regarding client-side security measures
    Sources used for this update (2)
    1. www.techtimes.co.uk — One ChatGPT Link Could Have Exposed Private Chats and ...
    2. privacyscrubber.com — ChatGPT macOS App Flaw: Sensitive Data Exposure & Local
    confidence 100%
  3. OpenAI Patches ChatGPT Share Link Flaw That Hijacked User Sessions

    OpenAI fixed a high-severity vulnerability in ChatGPT that allowed attackers to hijack user accounts and expose private conversations. Security researchers at Parameter Labs discovered that opening a malicious shared conversation link executed attacker-controlled JavaScript within an authenticated user session. This exploit allowed unauthorized parties to list victim conversations, read unshared private chats, and transmit stolen data to external servers. OpenAI implemented a patch for the flaw on October 26, following disclosure by Parameter Labs on October 28.

    Why it matters

    The vulnerability threatened users by requiring only a single click on a compromised shared link to compromise an active session. Because the exploit leveraged cross-frame messaging within signed-in environments, anyone who clicked a tainted link became a potential target for data theft. This incident highlights the security risks inherent in shared session features and third-party content rendering within artificial intelligence platforms.

    What is confirmed

    • A critical vulnerability in ChatGPT allowed a malicious shared conversation to run attacker-controlled JavaScript in another user's signed-in session.
    • The exploit allowed attackers to list private chats, read conversations never shared with the attacker, and send stolen data to an outside server.
    • OpenAI fixed the flaw on October 26.
    • Researchers at Parameter disclosed the vulnerability on October 28, 2026.

    Still unconfirmed

    • Apple also patched a vulnerability in macOS that allowed apps to access sensitive user data.

    What to watch next

    • Technical post-mortems or expanded scope disclosures from OpenAI or Parameter Labs.
    • Additional security audits targeting ChatGPT shared link and session handling features.
    Sources used for this update (7)
    1. cybersecuritynews.com — Critical ChatGPT Flaw Allowed Attackers to Hijack Accounts ...
    2. cyberinsider.co.uk — Patched ChatGPT sharing flaw exposed private conversations
    3. cyberpress.org — OpenAI Fixes High-Severity ChatGPT Flaw That Exposed Private ...
    4. labs.cloudsecurityalliance.org — One Click Into Someone Else’s ChatGPT: Cross-Frame Messaging ...
    5. aigovernance.com — Patched ChatGPT Account Takeover Exposes Shared Session as…
    6. note.com — ChatGPT shared links could be hijacked just by opening them ...
    7. aiweekly.co — OpenAI Patches ChatGPT Share-Link Flaw That Hijacked Sessions
    confidence 95%
  4. Flaw in ChatGPT's Mac App Allowed Hackers to Grab Sensitive Data

    A vulnerability in ChatGPT's Mac app and Apple macOS allowed hackers to access sensitive user data. Researchers at Parameter found a flaw in ChatGPT's shared link feature that let attackers run code in logged-in sessions, extracting private chat content. OpenAI fixed the flaw on October 26. Apple also patched a vulnerability in macOS that allowed apps to access sensitive user data.

    Why it matters

    This security issue compounds existing corporate challenges surrounding artificial intelligence applications on desktop systems. IT administrators face complications from employees using personal accounts at work and sharing sensitive data with AI tools. Previous vulnerabilities in desktop AI software and macOS have raised concerns about data protection.

    What is confirmed

    • A malicious ChatGPT share link could run an attacker's code in the logged-in session of anyone who opened it.
    • The ChatGPT flaw was reported through OpenAI's bug bounty program on October 18 and fixed on October 26.
    • Apple macOS contains a vulnerability tracked as CVE-2026-65379, allowing applications to gain access to sensitive user data.

    What to watch next

    • Further updates on corporate adoption of ChatGPT and AI tools
    • Details on the impact of the vulnerability on users
    Sources used for this update (4)
    1. thehackernews.com — A Shared ChatGPT Link Could Run Attacker Code in Your Session ...
    2. hamerintel.com — ChatGPT bug that let shared links steal private chats fixed…
    3. radar.offseq.com — CVE-2026-112441: macOS · An app may be able to access ...
    4. radar.offseq.com — CVE-2026-112577: macOS · An app may be able to access ...
    confidence 100%
  5. Mac App Flaw and macOS Vulnerabilities Threaten Data

    Apple macOS contains a vulnerability tracked as CVE-2026-65379, where an application may gain access to sensitive user data due to insufficient handling of secure input. This security flaw compounds existing corporate challenges surrounding artificial intelligence applications on desktop systems. Security researchers previously uncovered vulnerabilities in desktop AI software, including a local Mac application flaw that allowed malware to read chat logs and a GitHub integration flaw enabling secret-theft attacks. IT administrators also face complications from employees using personal accounts at work, while Samsung employees accidentally leaked trade secrets by sharing sensitive data with ChatGPT.

    Why it matters

    Artificial intelligence tools present complex data protection problems for enterprises as workers adopt personal accounts and desktop software. ChatGPT retains information to train models, creating internal compliance hurdles. Prior security disclosures demonstrate that local desktop software and integration flaws expose sensitive information to unauthorized access.

    What is confirmed

    • CVE-2026-65379 is a vulnerability in Apple macOS where an application may be able to access sensitive user data due to insufficient handling of secure input.
    • Security researchers previously uncovered vulnerabilities in desktop artificial intelligence software, including a local Mac application flaw that allowed malware to read chat logs and a GitHub integration flaw enabling secret-theft attacks.
    • Samsung employees previously leaked trade secrets by sharing sensitive data with ChatGPT, which retains information to train models.

    What to watch next

    • Official patches or vendor advisories addressing CVE-2026-65379 in Apple macOS.
    • Updates from OpenAI regarding security configurations for desktop applications.
    Sources used for this update (4)
    1. www.wired.com — Lily Hay Newman - WIRED
    2. techpilot.ai — Is ChatGPT Safe? Leaks, Privacy & 10 Fixes - TechPilot.ai
    3. radar.offseq.com — CVE-2026-65379: An app may be able to access sensitive user ...
    4. concentric.ai — 7 ChatGPT Security Risks to Focus on in 2026 | Concentric AI
    confidence 100%
  6. OpenAI Model Bug Discovery Follows Mac and Enterprise Security Risks

    Cybersecurity researchers using an OpenAI model discovered a bug in the widely used programming language PHP, drawing widespread internet attention. This incident compounds existing enterprise security challenges tied to artificial intelligence tools. Previous disclosures revealed that Samsung employees accidentally leaked trade secrets by sharing sensitive data with ChatGPT, which retains information to train models. Additionally, IT administrators face complications from employees using personal accounts at work, while security researchers previously uncovered vulnerabilities in desktop artificial intelligence software, including a local Mac application flaw that allowed malware to read chat logs and a GitHub integration flaw enabling secret-theft attacks.

    Why it matters

    Artificial intelligence tools present ongoing security risks for corporate networks and software development environments. Data sharing practices expose proprietary corporate information to model training pipelines, creating severe compliance and espionage risks. Meanwhile, local application flaws show that desktop artificial intelligence clients can expose user logs to malicious software.

    What is confirmed

    • Cybersecurity researchers using an OpenAI model found a bug in PHP, a widely used programming language.
    • Data shared with ChatGPT is retained and used to further train the model.
    • Samsung employees accidentally leaked top secret Samsung data by sharing information with ChatGPT.

    What to watch next

    • Official security patches or updates released for the PHP programming language bug discovered by researchers.
    • Additional enterprise policy changes regarding the use of personal artificial intelligence accounts at work.
    Sources used for this update (4)
    1. www.businessinsider.com — A Major Security Flaw Found Using ChatGPT Is Causing Internet ...
    2. londondaily.com — Whoops, Samsung workers accidentally leaked trade secrets via ...
    3. mediabias.news — Wired media bias and reliability — MediaBias News
    4. www.wfmz.com — 'Treasure trove of data': Grocery store websites ideal for ...
    confidence 90%
  7. Critical ChatGPT GitHub Integration Flaw Exposes Repositories to Theft

    A newly disclosed vulnerability in the ChatGPT GitHub integration allows attackers to conduct secret-theft attacks during routine repository reviews using manipulated instructions. Security researcher Sunil Tripathy published findings on October 15, 2026, detailing how prompt injection enables the vulnerability. This follows previous disclosures regarding security risks in desktop artificial intelligence software, including a local Mac application flaw discovered by Patrick Wardle that permitted malware to read chat logs. IT administrators are also facing challenges as employees use personal ChatGPT accounts at work, potentially exposing sensitive enterprise information.

    Why it matters

    The discovery highlights ongoing security challenges as artificial intelligence tools integrate deeply with developer platforms and source code repositories. Previous incidents involving the macOS client and web-based infostealer campaigns demonstrate that desktop and integrated AI software remain prime targets for threat actors. Organizations are increasingly forced to manage unauthorized personal account usage alongside platform-level vulnerabilities.

    What is confirmed

    • A critical vulnerability in the ChatGPT GitHub integration allows attackers to steal secrets from private repositories using manipulated instructions.
    • Security researcher Sunil Tripathy revealed the GitHub integration flaw in a report published on October 15, 2026.
    • Employees using personal ChatGPT accounts at work can expose sensitive company data even when organizations provide an approved Enterprise workspace.

    What to watch next

    • Official remediation steps or patches released by OpenAI for the GitHub connector vulnerability
    • Further technical disclosures regarding prompt injection attack vectors in developer tool integrations
    Sources used for this update (6)
    1. www.wired.com — WIRED - The Latest in Technology, Science, Culture and Business
    2. www.theguardian.com — OpenAI accuses employee who left for Apple of leaking plans ...
    3. arstechnica.com — OpenAI models hacked a crypto wallet and breached more websites
    4. cybersecuritynews.com — Critical ChatGPT GitHub Connector Flaw Could Allow Attackers ...
    5. www.bleepingcomputer.com — Employees using personal ChatGPT at work? Here's what IT ...
    6. blog.rankiteo.com — GitHub and OpenAI: Critical ChatGPT GitHub Connector Flaw ...
    confidence 100%
  8. macOS ChatGPT Flaw Exposed Local Data

    A security vulnerability in the official macOS ChatGPT client allowed local malware to read chat logs, steal conversation histories, and issue commands to connected services. Security researcher Patrick Wardle discovered the flaw, which was fixed on September 25. The incident underscores broader risks tied to desktop artificial intelligence software and local system access. Meanwhile, Apple has rolled out iOS 26.2 and macOS Tahoe 26.2 updates addressing more than 30 flaws, including four WebKit vulnerabilities co-discovered by OpenAI and Anthropic systems. Separate threat campaigns continue to abuse AI share links to deliver infostealer malware.

    Why it matters

    Desktop artificial intelligence applications require deep system integrations to automate tasks, creating lucrative targets for local attackers if permissions are mismanaged. The discovery of the ChatGPT macOS flaw highlights the danger of granting full disk access to third-party desktop assistants. Concurrently, broader operating system updates from Apple signal tightening scrutiny over how artificial intelligence systems interact with core components like WebKit.

    What is confirmed

    • Apple's iOS 26.2 and macOS Tahoe 26.2 updates patch 30-plus flaws, including four WebKit vulnerabilities co-discovered by OpenAI and Anthropic AI systems.
    • A flaw in the macOS ChatGPT client discovered by Patrick Wardle allowed local attackers to take control of the desktop application and access stored chat logs.
    • LLMShare discovered by Push Security abuses ChatGPT's share links on chatgpt.com to host fake outage pages that deliver infostealer malware to Windows and macOS users.

    Still unconfirmed

    • AI agents from OpenAI, Google, Meta and Anthropic have breached real systems during security tests.

    What to watch next

    • Further updates to macOS Full Disk Access privacy controls for desktop AI agents
    • Additional intelligence on the scope of the LLMShare infostealer campaign targeting chatgpt.com domain share links
    Sources used for this update (7)
    1. dailysecurityreview.com — Apple Patches 30+ Flaws as AI Systems Earn WebKit CVE Credit
    2. apple.gadgethacks.com — Apple macOS Full Disk Access AI Agents: What to Check Now
    3. www.tomsguide.com — AI agents are going rogue — here's what you need to know if ...
    4. dailysecurityreview.com — LLMShare Campaign Hosts Infostealer Downloads on ChatGPT's ...
    5. www.blackarrowcyber.com — Black Arrow Cyber Threat Intel Briefing 09 October 2026 ...
    6. www.macg.co — ChatGPT pouvait être détourné sur Mac pour lire vos ...
    7. www.strac.io — ChatGPT Security Risks in Enterprise: 2026 Guide to Data ...
    confidence 90%
  9. OpenAI Patches ChatGPT Mac App Flaw That Exposed Sensitive Data

    OpenAI fixed a security vulnerability in its official macOS ChatGPT client that could have permitted existing malware to access stored chat logs, steal conversation histories, and issue commands to connected services. Security researcher Patrick Wardle discovered the flaw, which allowed local attackers to take control of the desktop application. While the vulnerability has already been patched, the incident draws attention to the security risks facing desktop AI software and trusted local pipelines. Apple is also updating macOS privacy controls for Full Disk Access in response to growing privacy concerns surrounding desktop AI agents.

    Why it matters

    Desktop AI tools operate with high privileges and direct access to local user files, making them attractive targets for malicious actors. Security analysts emphasize that while much attention focuses on AI models executing cyberattacks, the software utilities themselves remain vulnerable to exploitation. The discovery prompted a re-evaluation of how desktop AI environments handle user data exfiltration and permissions on operating systems.

    What is confirmed

    • Security researcher Patrick Wardle found a vulnerability in the ChatGPT app for macOS.
    • OpenAI patched a security flaw in the ChatGPT macOS client that could let attackers access chat logs and sensitive application data.
    • The vulnerability could have allowed existing malware to access stored chats, issue commands, and reach connected services.
    • Apple is introducing stronger consent controls for Full Disk Access on macOS following privacy concerns over AI agents.

    What to watch next

    • Further updates from Apple regarding macOS privacy controls and Full Disk Access permissions for AI applications.
    • Additional security audits or disclosures regarding desktop AI client vulnerabilities.
    Sources used for this update (12)
    1. www.wired.com — 94 Best Prime Day Deals We’re Shopping This October (2026)
    2. techbora.com — A flaw in ChatGPT's Mac app could have let hackers grab ...
    3. ainave.com — ChatGPT Mac App Vulnerability: What the Flaw Exposed
    4. explore.n1n.ai — Security Vulnerability in ChatGPT Mac App Exposed Sensitive ...
    5. The420.in — ChatGPT Mac App Flaw Could Have Exposed User Chats to Hackers
    6. ai-brainer.com — ChatGPT macOS App Flaw Let Attackers Access Sensitive Data
    7. www.webanditnews.com — ChatGPT Mac App Flaw Exposed Chat Logs and Browser Sessions ...
    8. tetmo.com — ChatGPT Mac App Flaw Exposed AI Governance Risks - tetmo.com
    9. www.thetechedvocate.org — This OpenAI ChatGPT Mac App Security Flaw Could Have Exposed ...
    10. aistart.ai — ChatGPT Mac App Flaw Exposed Sensitive Data | AI News
    11. en.killbait.com — Security Flaw in ChatGPT's macOS App Allows Potential Data ...
    12. www.wionews.com — Apple changes Mac privacy controls after AI agents raise ...
    confidence 90%
  10. OpenAI Patches Security Flaw in ChatGPT macOS App

    OpenAI fixed a vulnerability in the ChatGPT macOS application on September 25, 2026, that could have allowed attackers to access sensitive user data. The flaw potentially exposed chat logs, files, and browser sessions. According to Objective-See, the vulnerability was a trivial trust bypass that could have granted hackers control over the app using a few lines of code. While much industry focus remains on AI agents conducting cyberattacks, this incident demonstrates that AI software itself is a vulnerable target for exploitation.

    Why it matters

    The vulnerability highlighted risks associated with granting AI applications deep system permissions on personal hardware. This occurs as developers race to create AI agents capable of executing tasks on behalf of users. The incident underscores a shift in risk from theoretical AI dangers to practical software vulnerabilities.

    What is confirmed

    • OpenAI patched a vulnerability in the ChatGPT macOS app that could have exposed user data.
    • The flaw potentially allowed access to chat logs, browser sessions, and files.
    • OpenAI reportedly fixed the issue on September 25, 2026.
    • The vulnerability could have been exploited with a few lines of code.

    Still unconfirmed

    • Objective-See found the macOS ChatGPT trust bypass.

    What to watch next

    • Confirmation of whether any users were actively exploited before the patch
    • Details on the specific nature of the trust bypass from OpenAI
    Sources used for this update (15)
    1. lifehacker.com — 10 Security Hacks Every Local AI User Should Know
    2. www.wired.com — A Flaw in ChatGPT’s Mac App Could Have Let Hackers Grab Sensitive Data
    3. WIRED — A Flaw in ChatGPT’s Mac App Could Have Let Hackers Grab Sensitive Data
    4. ForkLog — Vulnerability in ChatGPT macOS App Allowing Access to Conversations Fixed
    5. The420.in — ChatGPT Mac App Flaw Could Have Exposed User Chats to Hackers
    6. Mezha — ChatGPT for Mac could be hacked with just a few lines of code
    7. The Tech Buzz — ChatGPT Mac App Vulnerability Exposed User Data to Hackers
    8. bsky.app — @mcmillen.dev 님의 게시물
    9. flipboard.com — A Flaw in ChatGPT’s Mac App Could Have Let Hackers Grab Sensitive Data
    10. www.neoteo.com — ChatGPT Mac app vulnerability could expose chat data | NeoTeo
    11. www.deskofai.com — A Flaw in ChatGPT’s Mac App Could Have Let Hackers Grab ...
    12. flipboard.com — 10 Healthy Brain Habits To Keep Your Career Sharp This Fall
    confidence 90%
📊

Community Sentiment: How do you assess this situation?

Voice your perspective · Real-time aggregated sentiment from the Live Feeds community