A specific AI language model called Sec-PaLM strengthens Google Cloud with a separate cybersecurity suite.

At the RSA 2023 conference, Google Cloud announces its Security AI Workbench, a complete cybersecurity suite. AI is the magic word here that likes to vibrate. According to Google, the AI ​​language model is based on Sec-PaLM, a variant of Google’s own PaLM model. By bringing in information on research, software vulnerabilities, malware, behavior and associated indicators, the model must be able to make intelligent cybersecurity decisions.

Google Cloud Security AI Workbench comes with an entire book of AI-powered tools. Mandiant Threat Intelligence AI can be found inside to aggregate threats and respond quickly (Google bought Mandiant for $5.4 billion last year). VirusTotal, also owned by Google, also uses Sec-PaLM to analyze deceptive scripts and explain their behavior.

The full package also includes an assistant that guides Chronicle customers. Anyone using Google Security Command Center’s AI now gets a clear, less technical explanation of affected assets, risk analysis, privacy issues, and recommended countermeasures.

In its announcement, Google writes that Sec-PaLM does not react quickly to the hype surrounding ChatGPT and other AI models. “Sec-PaLM is based on years of AI research from Google and Deepmind, combined with the extensive expertise of our security teams. We are only now beginning to apply the power of generative AI to security and look forward to the advances we can make within the security community.”

Much announced, little available

The fact that Google Cloud also wants to combine security with AI is daring. According to TechCrunch, an AI language model still makes mistakes and they are sensitive to them prompt injection where the model reacts differently than expected.

Google has big ambitions with the suite, but very little is currently available. VirusTotal Code Insight is the first tool you can preview in Google Cloud Security AI Workbench. The cloud giant plans to test-roll the other services to “trusted testers” later this year before making them publicly available.

Sec-PaLM may also require tuning along the way. Google is currently sharing remarkably little information about the model. Hopefully they won’t be playing panic football like AI bot Bard is.