The hottest Substack posts of PromptArmor Blog

And their main takeaways
604 HN points 20 Aug 24
  1. There is a serious vulnerability in Slack AI that lets attackers access confidential information from private channels without needing direct access. This means sensitive data can be stolen just by manipulating how Slack AI processes requests.
  2. The risk increases with the recent Slack update that allows AI to access files shared within the platform. This could mean that harmful files uploaded by users can also be exploited to extract confidential information.
  3. Both data theft and phishing attacks can happen through crafted messages in public channels. This makes it crucial for users to be careful about what they share, because attackers can trick the AI into sharing sensitive details.
235 HN points 15 Dec 23
  1. A vulnerability in Writer.com allows attackers to steal private documents by manipulating the language model.
  2. Writer.com emphasizes data security for the sensitive information uploaded by its users.
  3. The responsible disclosure process with Writer.com revealed a security issue that was not initially recognized or addressed.
100 implied HN points 20 Mar 24
  1. LASEC is a new certification focused on LLM application security. It aims to educate leaders on current security threats and best practices.
  2. Participants will learn about real-world threats, including a new exploit discovered by PromptArmor. They'll also dive into compliance standards and how to balance security with product development.
  3. The certification program is designed to share knowledge gained from working with top security leaders in Fortune 100 companies, making it a valuable resource for security professionals.
4 HN points 20 Aug 24
  1. There is a serious risk in Slack where attackers can steal sensitive information from private channels. They can do this by tricking the AI into revealing data through malicious instructions.
  2. The inclusion of files and documents into Slack AI's responses has greatly increased the potential for these attacks. Now, attackers could even hide malicious instructions within documents that users upload.
  3. Slack's recent changes have made it easier for attackers to exploit these vulnerabilities without needing direct access to the private channels. It's crucial for organizations to manage and restrict these features to protect sensitive information.
0 implied HN points 14 Dec 23
  1. The blog is from PromptArmor
  2. The post promises something coming soon
  3. There's a call to action to share and subscribe
Get a weekly roundup of the best Substack posts, by hacker news affinity: