Top all-time · #2710
PromptArmor
Company blog · promptarmor.com/resourcessecurityai
PromptArmor is the company blog at promptarmor.com, writing about security and ai. Since 2024, 28 different Hacker News readers have linked it 59 times, most often “Claude Cowork Exfiltrates Files”. No feed was found, so read it on the site itself.
Too few sampled comments (9) to say how often HN recommends it.?
Why HN links it
HN readers link PromptArmor's write-ups of prompt injection and data exfiltration attacks on AI agents and coding tools, such as Google Antigravity, IBM Bob, Superhuman and Claude Cowork, as examples.
Summarised by Claude from a sample of 9 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
Assess and monitor risk from AI in vendors with novel intelligence on emerging threats. Stay ahead of AI-driven third-party risk with continuous monitoring and adaptive insight.
Most-linked posts?
- Claude Cowork Exfiltrates Files4 people · 4 mentions · first linked 2026-01-14
- Codex for Everything Exfiltrates Connected Data4 people · 4 mentions · first linked 2026-05-21
- Superhuman AI Exfiltrates Emails3 people · 3 mentions · first linked 2026-01-12
- Google Antigravity exfiltrates data via indirect prompt injection attack3 people · 3 mentions · first linked 2025-11-25
- IBM AI ('Bob') Downloads and Executes Malware2 people · 2 mentions · first linked 2026-01-08
- Data Exfiltration in Claude for Excel2 people · 2 mentions · first linked 2025-11-21
- Elastic Agentic SOC Vulnerable to Credential Theft2 people · 2 mentions · first linked 2026-09-28
- Ramifications of Using an Agent-in-the-Loop to Approve Commands2 people · 2 mentions · first linked 2026-06-03
- Remote Code Execution on a $1B Legal AI Tool2 people · 2 mentions · first linked 2025-12-12
- Unpatched Ollama Vulnerabilities: Phishing Overlays and Data Exfiltration2 people · 2 mentions · first linked 2026-05-29
What Hacker News says?
Here's prompt injection via a harness for example: https://www.promptarmor.com/resources/google-antigravity-exf... took me about 30 seconds to find. there's tons of them. of course it's an arms race, the models get better the harnesses get better the attackers get better but again: for the model itself, your prompt and third party data it reads are the same thing (input). as long as it is true, the attack will alway…
Comment linking /resources/google-antigravity-exfiltrates-data
on “Show HN: Auto Dismiss Mac OS "Allow <app> to find devices on local networks?"”Submitted Google Antigravity exfiltrates data via indirect prompt injection attack as a story · 768 points
HN discussion"IBM Bob is IBM’s new coding agent, currently in Closed Beta. " Promptarmor did a similar attack(1) on Google's Antigravity that is also a beta version. Since then, they added secure mode(2). These are still beta tools. When the tools are ready, I'd argue that they will probably be safer out of the box compared to a whole lot of users that just blindly copy-paste stuff from the internet, adding random dependencies w…
Comment linking /resources/google-antigravity-exfiltrates-data
on “IBM AI ('Bob') Downloads and Executes Malware”Submitted IBM AI ('Bob') Downloads and Executes Malware as a story · 49 points
HN discussionLiterally this from the past two weeks, a prompt injection attack that works on Superhuman, the AI email assistant application. https://www.promptarmor.com/resources/superhuman-ai-exfiltra... https://news.ycombinator.com/item?id=46592424
Comment linking /resources/superhuman-ai-exfiltrates-emails
on “Clawdbot - open source personal AI assistant”Submitted Claude Cowork Exfiltrates Files as a story · 45 points
HN discussionSource: https://www.promptarmor.com/resources/ibm-ai-(-bob-)-downloa...
Comment linking /resources/ibm-ai-(-bob-)-downloads-and-executes-malware
on “IBM's AI agent Bob easily duped to run malware, researchers show”Submitted Data Exfiltration in Claude for Excel as a story · 11 points
HN discussionI feel like prompt armor writes the exact same blog post for every agentic tool because they all suffer from the ignore previous instructions prompt injections. https://www.promptarmor.com/resources/claude-cowork-exfiltra... https://www.promptarmor.com/resources/google-antigravity-exf... https://promptarmor.substack.com/p/data-exfiltration-from-sl... https://www.promptarmor.com/resources/gpt-for-google-sheets-... ht…
Comment linking /resources/superhuman-ai-exfiltrates-emails
on “Atlassian Rovo Exfiltrates Data, Bypassing Controls”Submitted Open-weight models 88% more susceptible to cyber attacks as a story · 10 points
HN discussionCreator here. A few key differences: 1. from isolation pov, Matchlock launch Firecracker microvm with its own kernel, so you get hardware-level isolation rather than bubblewrap's seccomp/namespace approach, therefore a sandbox escape would require a VM breakout. 2. Matchlock intercepts and controls all network traffic by default, with deny-all networking and domain allowlisting. Bubblewrap doesn't provide this, whic…
Comment linking /resources/claude-cowork-exfiltrates-files
on “Matchlock: Linux-based sandboxing for AI agents”Submitted Data Exfiltration in Claude for Excel as a story · 9 points
HN discussion> apparently completely written by Claude Code https://www.promptarmor.com/resources/claude-cowork-exfiltra... > Claude Cowork Exfiltrates files That explains it
Comment linking /resources/claude-cowork-exfiltrates-files
on “Ask HN: Do you have any evidence that agentic coding works?”Submitted Screen Takeover Attack in AI Tool Acquired for $1B as a story · 7 points
HN discussion> I thought we might finally have a high profile prompt injection attack against a name-brand company we could point people to. These folks have found a bunch: https://www.promptarmor.com/resources But I guess you mean one that has been exploited in the wild?
Comment linking /resources
on “AI Agent Hacks McKinsey”Submitted Remote Code Execution on a $1B Legal AI Tool as a story · 6 points
HN discussion> LLMs agents are not exploiting kernels to get out of the sandbox. You can't assume that. Attackers with LLMs have enough capabilities to engineer them to build exploits for kernel vulnerabilities [0] or to bypass sandboxes to exfiltrate data [0] in covert ways. It is completely possible to craft a chained attack for an agent to bypass sandboxes even with or without a kernel exploit. From [0] and [1] [0] https://se…
Comment linking /resources/claude-cowork-exfiltrates-files
on “Running Claude Code dangerously (safely)”Submitted Unpatched Ollama Vulnerabilities: Phishing Overlays and Data Exfiltration as a story · 5 points
HN discussionSubmitted HuggingFace Chat Exfiltrates Data as a story · 4 points
HN discussionSubmitted A One Pixel Image Can Leak Your Data in HuggingChat as a story · 4 points
HN discussionSubmitted Microsoft Copilot Cowork Controlled by Attacker, Bypassing Sandbox as a story · 3 points
HN discussionSubmitted Attacker Takes over Zoom AI as a story · 3 points
HN discussionSubmitted Microsoft Copilot Cowork Exfiltrates Files as a story · 3 points
HN discussionSubmitted Codex for Everything Exfiltrates Connected Data as a story · 3 points
HN discussionSubmitted GitHub Copilot CLI Downloads and Executes Malware as a story · 3 points
HN discussionSubmitted OpenAI Codex PSA on Malicious Config Files as a story · 3 points
HN discussionSubmitted Superhuman AI Exfiltrates Emails as a story · 3 points
HN discussionSubmitted Notion AI: Unpatched Data Exfiltration as a story · 3 points
HN discussionSubmitted $1B AI Tool Screen Takeover Attack as a story · 3 points
HN discussionSubmitted Hijacking Copilot Cowork's AI Gateway to Bypass Sandboxing and Exfiltrate Files as a story · 2 points
HN discussion
