Fecha de publicación: 12/06/2024 Fuente: Noticias ICEX Lugar: Threat Intelligence Threat IntelligenceGenerate vulnerability intelligence reports in minutes with AI Actions Synthesize and analyze vulnerabilities to extract data and triage risks faster 15 sec-summaryWe previously discussed how AI Actions can synthesize multiple articles to extract insights and create customized intelligence reports in minutes. In this post, we focus on the vulnerability analysis use case and shift to more advanced reporting prompts that can generate specific and targeted output to meet your intelligence needs.With AI Actions, you can:Create vulnerability briefs showing the key attributes of the vulnerabilities to accelerate your analysis and recommendations.Customize your vulnerability reporting with unique and specific prompts to generate tables, charts, or formatted reports using some of our best practices for engineering prompts.Save your vulnerability AI Actions for reuse, delivering a consistent reporting format for your stakeholders.AI Actions help vulnerability managers accelerate their analysis and triaging of vulnerabilities by synthesizing and extracting data to generate vulnerability reports aligned with stakeholders’ needs.Start free trialChallenges of using AI in cybersecurityThe old mantra of “garbage in, garbage out” holds true in the emerging world of Generative AI. If the LLMs are trained on bad data, your answers will likely be poor. In addition, most GenAI tools are trained on datasets that don’t reflect emerging news - a critical gap for cyber threat intelligence teams.Feedly for Threat Intelligence’s AI Actions can be trusted because the output is limited to the input you provide. That is, the articles you select for analysis represent the dataset for the response. If the answer isn’t contained in the articles you select, AI Actions will inform you, rather than providing a potential hallucination. In addition, the results contain inline citations, making answers verifiable, transparent, and allowing you to dive deeper into the source the information was pulled from.The garbage in, garbage out rule doesn’t just apply to the data; it applies to the questions you ask the AI. If they are worded poorly or don’t specify the output or the tone you want, it’s hard to get consistently good answers that meet your needs. In cybersecurity, you might have a defined structure for data or intelligence reports. An answer that doesn’t follow that structure means more work for you to reformat or rewrite it.In this blog, we’ll use the vulnerability management use case to illustrate how you can use simple questions to well-engineered prompts to create the precise output you want.Create a vulnerability brief showing key attributes of the vulnerabilitiesVulnerability analysts must quickly understand and prioritize new or specific CVEs (Common Vulnerabilities and Exposures). A typical task is to quickly synthesize, understand, and triage the risks to your organization related to a vendor advisory such as Microsoft’s Patch Tuesday releases. You’ll also likely want to get a complete picture of the potential risks by looking at sources beyond the vendor advisory.With AI Actions, you can select the relevant articles you want to synthesize, analyze, and ask Feedly to create a vulnerability brief.The output of this simple prompt is useful, but it relied heavily on the underlying LLM to interpret the question and generate a response. It may not have aligned with your expectations. What if you wanted to see if there are CVSS scores, available patches, or exploits for each of the vulnerabilities?Let’s try a more advanced prompt requesting that data be presented in a specified format that closely aligns with the desired end report. In this AI Action prompt, we detail the purpose and important things we want the AI to look for and clearly define the desired output in more certain terms.Start free trialCustomize your vulnerability reporting to generate tables, charts, or formatted reports using our prompt engineering best practicesStart with the prompt libraryTo give you a head start, AI Actions include crafted prompts prepared by Feedly. These prompts were developed in collaboration with our beta customers. We worked to understand their needs, experimented with diverse outputs, and finally arrived at an approved result.That said, they might not be perfect for you. Fortunately, you can see and copy the prompts in the prompt library and use them to construct new prompts that meet your specific requirements.Write customized promptsSometimes, you just want to summarize content for your own consumption. Other times, you may need very specific analysis or to extract unstructured data embedded within an article into something usable by your stakeholders. Creating a good prompt can save you a lot of time rewriting, translating, or reformatting the output.Here’s a simple pie chart of vulnerabilities by criticality that could be used in executive reports.Crafting a prompt that’s useful requires a bit of art and science. Let’s start with the “science” in the form of best practices we’ve found:1. Provide the proper context: Tell the AI your your role and the intended audience. Stating that you are a cybersecurity analyst preparing a summary report for the CISO will help the AI provide an appropriate response.2. State the type of response you want: Do you want a summary, a short answer, a bulleted list, a table, a chart, etc.? Set character limits to help keep report sections concise.3. Specify the voice and tone: Should it be formal, technical, or informative?4. Clarify the format: Do you want the output in English, markdown, JSON, or some other format?5. Structure your prompts: Instead of using large text blocks, use short sentences, line breaks, and bullet points to structure your prompts clearly.And a little art:6. Experiment to see what works: Try different tones, outputs, and formats to see how the questions alter the output. Decide what works for you and your stakeholders and reuse those instructions in future prompts.Clear and well-structured prompts are easier to read, understand, and debug. As you create more complex prompts or share them with others, maintaining a clear structure and format ensures that prompts can be easily modified and adapted to new situations.By adopting these best practices from the start, you will develop habits and a mindset focused on effective, clear, and maintainable prompts. This will not only improve your team’s immediate results, but also equip you to create, iterate, and refine your prompts more efficiently over time.We encourage you to follow our guide for writing complex and effective prompts to drive consistency across your team.Make sure to save your prompts to reuse in the futureFinally, you may decide to change your prompt from time to time. You don’t have to start from scratch. You can select the saved prompt from your library, click on the (…), edit and resave it.In short...Quickly analyzing new vulnerability articles and advisories has never been easier. AI Actions can help you synthesize the key points across multiple articles and turn them into useful and consistent reports customized to your stakeholders’ needs.Using advanced prompt techniques can significantly improve your analysis and reporting capabilities, saving you a ton of time and making you look like a superstar. However, it’s important to use good prompt hygiene to provide clear instructions to the LLM, simplify debugging, and improve long-term maintainability. For more information about how to craft interesting prompts to get the most out of AI Actions, check out the documentation.Try AI ActionsQuickly synthesize content and create trusted, verifiable reports.START FREE TRIAL