• Secure AI
  • Posts
  • Secure AI #1: Welcome to Secure AI

Secure AI #1: Welcome to Secure AI

Author

Charles Gillman
July 23, 2023

Welcome to Secure AI, where we discuss the intersection of Cybersecurity, Privacy and IT Governance, Risk & Compliance.

In this newsletter…

  • AI Security News - Latest news & news you may have missed

  • Vendor AI Security News - AI news from the vendor world

  • Article of the Week - AI Just Changed Phishing

  • Recommended Reading - Deeper dives into AI & Security

AI Security News

AI Solutions for IoT Security: Safeguarding Low-Resource Devices with AI

- AI solutions are revolutionizing IoT security by detecting anomalies based on behavior, not just rules. - Automation and continuous monitoring powered by AI ensure timely detection and response to…

medium.com/@multiplatform.ai/ai-solutions-for-iot-security-safeguarding-low-resource-devices-with-ai-f337a5a88c4

ChatGPT AI Chatbot: Risks to Your Intellectual Property (IP) and How to Protect It

As businesses and individuals rely more and more on chatbots like ChatGPT, it is crucial to ensure that your intellectual property (IP) is protected.

blog.mindset.ai/blog/chatgpt-ai-chatbot-risks-to-your-intellectual-property-ip-and-how-to-protect-it?utm_campaign=625ee0e24341d4000106058b&utm_content=643e92101f505f0001197831&utm_medium=smarpshare&utm_source=linkedin

Concordium & 2021.AI: The Convergence of AI and Blockchain Revolutionizing Trust & Security

In the rapidly evolving world of technology, two of the most exciting fields are artificial intelligence (AI) and blockchain. The…

medium.com/@davidmygind_18067/concordium-2021-ai-the-convergence-of-ai-and-blockchain-revolutionizing-trust-security-d48139dab9f7

Vendor AI News

Splunk introduces Splunk AI, a suite of AI-driven solutions to enhance security and observability

- Splunk introduces Splunk AI, a suite of AI-driven solutions to enhance security and observability. - Splunk AI Assistant offers an interactive chat experience and expands understanding of the…

medium.com/@multiplatform.ai/splunk-introduces-splunk-ai-a-suite-of-ai-driven-solutions-to-enhance-security-and-observability-5dc5b3db6b5f

Article of the Week - AI Just Changed Phishing

We’ve trained users to spot phishing emails that make it through our phishing filters to look out for poor spelling and poor grammar. Now that cybercriminals have access to Bard, ChatGPT and other Large Language Models (LLMs), they can craft very realistic-looking phishing emails without spelling or grammatical errors.

The cybercriminals have also created a tool based on the GPTJ LLM, called WormGPT. WormGPT is designed to create realistic phishing and Business Email Compromise (BEC) emails.

Conversely, Google is releasing Help Me Write, and Microsoft is releasing Office Copilot, which will use their respective LLMs to help users create or reply to emails. So now we have AI generating both phishing and legitimate emails, making it even harder for users to spot scams. Cyber defenders will have to rely more on email filtering tools.

BEC emails will also be even more convincing, especially as LLMs can take input, such as a blog post by a CFO for example, and create output in the same writing style as the input. The result will be a very convincing BEC email.

The odds are now starting to favour the cybercriminals where they have started using AI to their advantage.

Recommended Reading

LLM Censorship: A Machine Learning Challenge or a Computer Security Problem?

Large language models (LLMs) have exhibited impressive capabilities in comprehending complex instructions. However, their blind adherence to provided instructions has led to concerns regarding risks of malicious use. Existing defence mechanisms, such as model fine-tuning or output censorship using LLMs, have proven to be fallible, as LLMs can still generate problematic responses. Commonly employed censorship approaches treat the issue as a machine learning problem and rely on another LM to detect undesirable content in LLM outputs. In this paper, we present the theoretical limitations of such semantic censorship approaches.

arxiv.org/abs/2307.10719

Remember AI won’t take your job, but someone that knows how to leverage AI probably will