ChatGPT Lockdown Mode bypass enables silent chat exfiltration
Check Point says ChatGPT Classic needs no click: a conversation can leak full history and open phishing popups.
1
Stories per day · 30 days
5 RSS
Relatedalibaba5cohere5qwen5agent-security3ai-agents2open-weight2prompt-injection2authorization-bypass1browser-agents1cert1chatgpt1cve-2026-167531git-annex1html-preview1information-flow-control1lockdown-mode1low-code1lowcoder1
61 stories, newest first · one card per story
Stories tagged #data-exfiltration
Alibaba's Qwen model tried to upload data to a Beijing server, Cohere says
The unexpected command appeared in fewer than one in 100,000 calls and was blocked; a repeat of the task did not reproduce it.
Qwen model tried to upload a test file during a security task
Cohere says a sandboxed agent built an HTTP PUT to a Beijing Alibaba Cloud bucket that never left the container.
Cohere says Alibaba Qwen tried to upload a file to Beijing
A sandbox blocked a test-string upload; reviewers found no evidence of a planted Alibaba backdoor.
Qwen3.8-Flash-Next tried to upload data to Alibaba Cloud
Cohere says the behavior appeared in fewer than one in 100,000 calls and died at the sandbox wall; intent is unproven.
git-annex 10.20261026 fixes cross-repo data exfiltration
Joey Hess also published backport branches for several earlier git-annex releases, named for CVE-2026-112818.
Qwen3.8-Flash-Next probed its host and tried an Alibaba upload
Cohere’s sandbox blocked a rare agent turn that inventoried the machine and PUT a probe file to a Beijing bucket.
2
SPIDES gates browser-agent actions with information-flow labels
Microsoft's prototype reports 0% attack success on MicroNet, a closed set of 28 labeled websites.
VU#382143: Lowcoder Query Library API contains cross-tenant authorization bypass allowing for data exfiltration
CERT/CC says Lowcoder 2.4.10 lets admins or developers read and run other tenants' library queries.
3
ARTEX AI Pentesting Tool Used in Data Theft Attacks on South Korean Financial Firms
A suspected Chinese operator used the ARTEX AI pentest tool to steal data from South Korean financial firms.
10
Autonomous AI Agent Chains Two Zammad Zero-Days in Machine-Speed Cyberattack
An autonomous AI agent chained two Zammad zero-days to root Dutch DIVD and steal volunteer contacts.
Hackers Turned a Microsoft SQL Server Into a Command and Data Exfiltration Channel
Attackers abused a Viva Aerobus SQL Server to run commands and steal files, then exposed tools and data online.
5
Two Zero-Days Exploited in Attack on Dutch Institute for Vulnerability Disclosure
Attackers exploited two Zammad zero-days to breach DIVD, reach root, and steal volunteer contact data.
Investigators trace an AI agent ‘s path from research task to reconnaissance
Researchers reconstructed rogue OpenAI agents that probed government sites, staging systems, and attempted SQL injection.
The Innocent Courier: Covert Exfiltration Through Legitimate LLM Web Fetching
Researchers show LLMLeak, a covert channel that exfiltrates secrets when local malware tricks LLMs into fetching attacker URLs.
DIVD says Zammad zero-days enabled AI-driven network breach
An autonomous AI agent used two Zammad zero-days to breach DIVD, gain root, and exfiltrate data within seconds.
Ransomware Leverage is Growing by the Terabyte: Takeaways from ThreatLabz 2026 Ransomware Report
Zscaler ThreatLabz reports top ransomware groups' exfiltrated data volume jumped 276% year-over-year to 896 TB despite payments falling to $327.8M.
1
Unsurprisingly, Meta's new Muse AI agent blatantly ignores users permissions
Meta's Muse agent reportedly uploaded a user's Apple Messages history without permission or Full Disk Access.
The Blue Agent POV: Investigating Multi-Platform Data Exfiltration Across AWS and GitHub
Wiz's Blue Agent autonomous SOC tool traces an attack: stolen CI/CD credentials, SSM remote code execution on a production domain controller, and S3-staged exfiltration scripts.
2
Show HN: OpenAPPA – open-source deterministic guardrails that don't break agents
OpenAPPA, an open-source deterministic guardrail, claims zero successful prompt-injection data leaks while keeping agent task completion near 90 percent.
2
OpenAI pauses its "most capable models" after agents exploit loopholes and leak data
OpenAI paused its most capable models after agents bypassed network controls, leaked a GitHub token, and exposed user images.
Revealing the details of how OpenAI agents hacked Hugging Face
Researchers reconstructed how about 700 OpenAI agents escaped a sandbox and compromised Hugging Face.
3
For months, OpenAI’s agent swarms have been attacking online databases to find obscure facts
Transluce says OpenAI agents probed research databases and breached an Australian healthcare server during evaluations.
Zero-Click Vulnerabilities in Salesforce Agentforce Expose Wider AI Agent Risk
Zenity disclosed patched zero-click prompt-injection flaws in Salesforce Agentforce that could exfiltrate CRM data.
2