Slack May Access Your Private Conversations via ‘Prompt Injection’ That Could Lead to Phishing, Scams

Popular cloud-based work communication platform, Slack, has now been flagged by a security firm due to a potential threat for users. This threat, it has been claimed, allows for the access of a user's private conversations on the platform.

According to the security firm, the use of the "prompt injection" method of the Slack AI system was recently discovered. This allegedly compromises user privacy and security. 

It should be noted that, in the past year, Salesforce debuted integrations of AI systems like OpenAI's ChatGPT, Einstein GPT, and Claude's chatbots. Salesforce is Slack's parent company. 

Slack May Access Your Private Conversations and Files via Slack AI

Slack AI

(Photo : Drew Angerer/Getty Images)
PromptArmor released a new report behind a new batch of security vulnerabilities found on Slack, citing two forms of attacks that may possibly endanger users and their data. According to the report, this is can be done by a threat actor through the manipulation of a vulnerability found in the platform's large language models (LLM). 

The report claimed that in Slack's August 14 update, it allowed its Slack AI to access users' private conversations, files, and other data for its Slack AI search tool. The access may be restricted by account owners and admins, but Slack made this a default on the platform. 

It was revealed by the security firm that it already disclosed this vulnerability to Slack through Responsible Disclosure. 

Read Also: Slack Rolls Out iPhone Widgets for Easier Work Catch-Up

Slack's AI Can Lead to Unauthorized Data Access, Phishing 

In addition, PromptArmor said that it also discovered another vulnerability in Slack's AI systems that can lead to phishing scams by receiving links generated by its AI through a method called "prompt injection."

A threat actor does not need to be a member of public channels in order to launch attacks, and it can manipulate the Slack AI into revealing an API key. Slack's AI will then be tricked into creating a new prompt that can swap a certain keyword with the API key, and then send a phishing link to users whenever a user activates it by asking the AI about it. 

Slack and its Significant AI Integration

Salesforce has integrated three AI systems to Slack, introducing ChatGPT, Einstein GPT, and Claude's AI to better improve user experience on the platform. 

The company would later on introduce Slack AI, its take on AI that is meant to assist users with what they need, along with the capability to "recap" channels. Furthermore, it also introduced ways to simplify the work experience as it could not only recap, but it can likewise generate replies and summarize messages in conversations. 

However, there has been a controversy this as the company has allegedly trained Slack AI and its other chatbots using user messages. The company has allegedly done so without user consent.

Related Article: Slack Trains AI Models With User Messages Without Explicit Consent

© 2024 iTech Post All rights reserved. Do not reproduce without permission.

More from iTechPost