Episode 502 - Azure Open AI and Security
Azure Open AI is widely used in industry but there are number of security aspects that must be taken into account when using the technology. Luckily for us, Audrey Long, a Software Engineer at Microsoft, security expert and renowned conference speaker, gives us insights into securing LLMs and provides various tips, tricks and tools to help developers use these models safely in their applications.   Media file: https://azpodcast.blob.core.windows.net/episodes/Episode502.mp3 YouTube: https://youtu.be/64Achcz97PI Resources: AI Tooling: Azure AI Tooling Announcing new tools in Azure AI to help you build more secure and trustworthy generative AI applications | Microsoft Azure Blog Prompt Shields to detect and block prompt injection attacks, including a new model for identifying indirect prompt attacks before they impact your model, coming soon and now available in preview in Azure AI Content Safety.  Groundedness detection to detect “hallucinations” in model outputs, coming soon.  Safety system messagesto steer your model’s behavior toward safe, responsible outputs, coming soon. Safety evaluations to assess an application’s vulnerability to jailbreak attacks and to generating content risks, now available in preview.   Risk and safety monitoring to understand what model inputs, outputs, and end users are triggering content filters to inform mitigations, coming soon, and now available in preview in Azure OpenAI Service. AI Defender for Cloud AI Security Posture Management AI security posture management (Preview) - Microsoft Defender for Cloud | Microsoft Learn AI Workloads Enable threat protection for AI workloads (preview) - Microsoft Defender for Cloud | Microsoft Learn        AI Red Teaming Tool Announcing Microsoft’s open automation framework to red team generative AI Systems | Microsoft Security Blog AI Development Considerations:   AI Assessment from Microsoft Conduct an AI assessment using Microsoft’s Responsible AI Impact Assessment Template Responsible AI Impact Assessment Guide for detailed instructions Microsoft Responsible AI Processes Follow Microsoft’s Responsible AI principles: fairness, reliability, safety, privacy, security, inclusiveness, transparency, and accountability Utilize tools like the Responsible AI Dashboard for continuous monitoring and improvement Define Use Case and Model Architecture Determine the specific use case for your LLM Design the model architecture, focusing on the Transformer architecture   Content Filtering System How to use content filters (preview) with Azure OpenAI Service - Azure OpenAI | Microsoft Learn Azure OpenAI Service includes a content filtering system that works alongside core models, including DALL-E image generation models. This system uses an ensemble of classification models to detect and prevent harmful content in both input prompts and output completions The filtering system covers four main categories: hate, sexual, violence, and self-harm Each category is assessed at four severity levels: safe, low, medium, and high Additional classifiers are available for detecting jailbreak risks and known content for text and code. JailBreaking Content Filters Red Teaming the LLM Plan and conduct red teaming exercises to identify potential vulnerabilities Use diverse red teamers to simulate adversarial attacks and test the model’s robustness Microsoft AI Red Team building future of safer AI | Microsoft Security Blog Create a Threat Model with OWASP Top 10 owasp.org/www-project-top-10-for-large-language-model-applications/assets/PDF/OWASP-Top-10-for-LLMs-2023-slides-v1_1.pdf Develop a threat model and implement mitigations based on identified risks   Other updates: Los Angeles Azure Extended Zones Carbon Optimization App Config Ref GA OS SKU In-Place Migration for AKS Operator CRD Support with Azure Monitor Managed Service Azure API Center Visual Studio Code Extension Pre-release Azure API Management WordPress Plugin Announcing a New OpenAI Feature for Developers on Azure
August 15, 2024
Audio is streamed directly from the publisher (azpodcast.blob.core.windows.net) as published in their RSS feed. Play Podcasts does not host this file. Rights-holders can request removal through the copyright & takedown page.
Show Notes
<p>Azure Open AI is widely used in industry but there are number of security aspects that must be taken into account when using the technology. Luckily for us, Audrey Long, a Software Engineer at Microsoft, security expert and renowned conference speaker, gives us insights into securing LLMs and provides various tips, tricks and tools to help developers use these models safely in their applications.</p>
<p> <img src="http://azpodcast.azurewebsites.net/image.axd?picture=%2f2024%2f08%2faudrey.jpg" alt="" /></p>
<p>Media file: <a href="https://azpodcast.blob.core.windows.net/episodes/Episode502.mp3">https://azpodcast.blob.core.windows.net/episodes/Episode502.mp3</a></p>
<p>YouTube: <a href="https://youtu.be/64Achcz97PI">https://youtu.be/64Achcz97PI</a></p>
<p>Resources:</p>
<p><strong>AI Tooling:</strong></p>
<ol>
<li><strong>Azure AI Tooling </strong><a id="menuramn" class="fui-Link ___1rxvrpe f2hkw1w f3rmtva f1ewtqcl fyind8e f1k6fduh f1w7gpdv fk6fouc fjoy568 figsok6 f1hu3pq6 f11qmguv f19f4twv f1tyq0we f1g0x7ka fhxju0i f1qch9an f1cnd47f fqv5qza f1vmzxwi f1o700av f13mvf36 f1cmlufx f9n3di6 f1ids18y f1tx3yz7 f1deo86v f1eh06m1 f1iescvh fhgqx19 f1olyrje f1p93eir f1nev41a f1h8hb77 f1lqvz6u f10aw75t fsle3fq f17ae5zn" title="https://azure.microsoft.com/en-us/blog/announcing-new-tools-in-azure-ai-to-help-you-build-more-secure-and-trustworthy-generative-ai-applications/" href="https://azure.microsoft.com/en-us/blog/announcing-new-tools-in-azure-ai-to-help-you-build-more-secure-and-trustworthy-generative-ai-applications/" rel="noreferrer noopener" target="_blank">Announcing new tools in Azure AI to help you build more secure and trustworthy generative AI applications | Microsoft Azure Blog</a>
<ul>
<li><a id="menuramp" class="fui-Link ___1rxvrpe f2hkw1w f3rmtva f1ewtqcl fyind8e f1k6fduh f1w7gpdv fk6fouc fjoy568 figsok6 f1hu3pq6 f11qmguv f19f4twv f1tyq0we f1g0x7ka fhxju0i f1qch9an f1cnd47f fqv5qza f1vmzxwi f1o700av f13mvf36 f1cmlufx f9n3di6 f1ids18y f1tx3yz7 f1deo86v f1eh06m1 f1iescvh fhgqx19 f1olyrje f1p93eir f1nev41a f1h8hb77 f1lqvz6u f10aw75t fsle3fq f17ae5zn" title="https://aka.ms/promptshields-techblog" href="https://aka.ms/promptshields-techblog" rel="noreferrer noopener" target="_blank"><strong>Prompt Shields</strong></a> to detect and block prompt injection attacks, including a new model for identifying indirect prompt attacks before they impact your model, coming soon and now available in preview in Azure AI Content Safety. </li>
<li><a id="menuramr" class="fui-Link ___1rxvrpe f2hkw1w f3rmtva f1ewtqcl fyind8e f1k6fduh f1w7gpdv fk6fouc fjoy568 figsok6 f1hu3pq6 f11qmguv f19f4twv f1tyq0we f1g0x7ka fhxju0i f1qch9an f1cnd47f fqv5qza f1vmzxwi f1o700av f13mvf36 f1cmlufx f9n3di6 f1ids18y f1tx3yz7 f1deo86v f1eh06m1 f1iescvh fhgqx19 f1olyrje f1p93eir f1nev41a f1h8hb77 f1lqvz6u f10aw75t fsle3fq f17ae5zn" title="https://aka.ms/groundednessdetection-techblog" href="https://aka.ms/groundednessdetection-techblog" rel="noreferrer noopener" target="_blank"><strong>Groundedness detection</strong></a> to detect “hallucinations” in model outputs, coming soon. </li>
<li><a id="menuramt" class="fui-Link ___1rxvrpe f2hkw1w f3rmtva f1ewtqcl fyind8e f1k6fduh f1w7gpdv fk6fouc fjoy568 figsok6 f1hu3pq6 f11qmguv f19f4twv f1tyq0we f1g0x7ka fhxju0i f1qch9an f1cnd47f fqv5qza f1vmzxwi f1o700av f13mvf36 f1cmlufx f9n3di6 f1ids18y f1tx3yz7 f1deo86v f1eh06m1 f1iescvh fhgqx19 f1olyrje f1p93eir f1nev41a f1h8hb77 f1lqvz6u f10aw75t fsle3fq f17ae5zn" title="https://learn.microsoft.com/en-us/azure/ai-services/openai/concepts/system-message" href="https://learn.microsoft.com/en-us/azure/ai-services/openai/concepts/system-message" rel="noreferrer noopener" target="_blank"><strong>Safety system messages</strong></a>to steer your model’s behavior toward safe, responsible outputs, coming soon.</li>
<li><a id="menuramv" class="fui-Link ___1rxvrpe f2hkw1w f3rmtva f1ewtqcl fyind8e f1k6fduh f1w7gpdv fk6fouc fjoy568 figsok6 f1hu3pq6 f11qmguv f19f4twv f1tyq0we f1g0x7ka fhxju0i f1qch9an f1cnd47f fqv5qza f1vmzxwi f1o700av f13mvf36 f1cmlufx f9n3di6 f1ids18y f1tx3yz7 f1deo86v f1eh06m1 f1iescvh fhgqx19 f1olyrje f1p93eir f1nev41a f1h8hb77 f1lqvz6u f10aw75t fsle3fq f17ae5zn" title="https://aka.ms/safety-evals-blog" href="https://aka.ms/Safety-Evals-Blog" rel="noreferrer noopener" target="_blank"><strong>Safety evaluations</strong></a> to assess an application’s vulnerability to jailbreak attacks and to generating content risks, now available in preview. </li>
<li><a id="menuran1" class="fui-Link ___1rxvrpe f2hkw1w f3rmtva f1ewtqcl fyind8e f1k6fduh f1w7gpdv fk6fouc fjoy568 figsok6 f1hu3pq6 f11qmguv f19f4twv f1tyq0we f1g0x7ka fhxju0i f1qch9an f1cnd47f fqv5qza f1vmzxwi f1o700av f13mvf36 f1cmlufx f9n3di6 f1ids18y f1tx3yz7 f1deo86v f1eh06m1 f1iescvh fhgqx19 f1olyrje f1p93eir f1nev41a f1h8hb77 f1lqvz6u f10aw75t fsle3fq f17ae5zn" title="https://aka.ms/safety-monitoring-blog" href="https://aka.ms/Safety-Monitoring-Blog" rel="noreferrer noopener" target="_blank"><strong>Risk and safety monitoring</strong></a> to understand what model inputs, outputs, and end users are triggering content filters to inform mitigations, coming soon, and now available in preview in Azure OpenAI Service.</li>
</ul>
</li>
<li><strong>AI Defender for Cloud</strong>
<ul>
<li>AI Security Posture Management <a id="menuran3" class="fui-Link ___1rxvrpe f2hkw1w f3rmtva f1ewtqcl fyind8e f1k6fduh f1w7gpdv fk6fouc fjoy568 figsok6 f1hu3pq6 f11qmguv f19f4twv f1tyq0we f1g0x7ka fhxju0i f1qch9an f1cnd47f fqv5qza f1vmzxwi f1o700av f13mvf36 f1cmlufx f9n3di6 f1ids18y f1tx3yz7 f1deo86v f1eh06m1 f1iescvh fhgqx19 f1olyrje f1p93eir f1nev41a f1h8hb77 f1lqvz6u f10aw75t fsle3fq f17ae5zn" title="https://learn.microsoft.com/en-us/azure/defender-for-cloud/ai-security-posture" href="https://learn.microsoft.com/en-us/azure/defender-for-cloud/ai-security-posture" rel="noreferrer noopener" target="_blank">AI security posture management (Preview) - Microsoft Defender for Cloud | Microsoft Learn</a></li>
<li>AI Workloads <a id="menuran5" class="fui-Link ___1rxvrpe f2hkw1w f3rmtva f1ewtqcl fyind8e f1k6fduh f1w7gpdv fk6fouc fjoy568 figsok6 f1hu3pq6 f11qmguv f19f4twv f1tyq0we f1g0x7ka fhxju0i f1qch9an f1cnd47f fqv5qza f1vmzxwi f1o700av f13mvf36 f1cmlufx f9n3di6 f1ids18y f1tx3yz7 f1deo86v f1eh06m1 f1iescvh fhgqx19 f1olyrje f1p93eir f1nev41a f1h8hb77 f1lqvz6u f10aw75t fsle3fq f17ae5zn" title="https://learn.microsoft.com/en-us/azure/defender-for-cloud/ai-onboarding" href="https://learn.microsoft.com/en-us/azure/defender-for-cloud/ai-onboarding" rel="noreferrer noopener" target="_blank">Enable threat protection for AI workloads (preview) - Microsoft Defender for Cloud | Microsoft Learn</a> </li>
</ul>
</li>
<li><strong> AI Red Teaming Tool</strong>
<ul>
<li><a id="menuran7" class="fui-Link ___1rxvrpe f2hkw1w f3rmtva f1ewtqcl fyind8e f1k6fduh f1w7gpdv fk6fouc fjoy568 figsok6 f1hu3pq6 f11qmguv f19f4twv f1tyq0we f1g0x7ka fhxju0i f1qch9an f1cnd47f fqv5qza f1vmzxwi f1o700av f13mvf36 f1cmlufx f9n3di6 f1ids18y f1tx3yz7 f1deo86v f1eh06m1 f1iescvh fhgqx19 f1olyrje f1p93eir f1nev41a f1h8hb77 f1lqvz6u f10aw75t fsle3fq f17ae5zn" title="https://www.microsoft.com/en-us/security/blog/2024/02/22/announcing-microsofts-open-automation-framework-to-red-team-generative-ai-systems/?msockid=18290294b2206337069616d3b30c629e" href="https://www.microsoft.com/en-us/security/blog/2024/02/22/announcing-microsofts-open-automation-framework-to-red-team-generative-ai-systems/?msockid=18290294b2206337069616d3b30c629e" rel="noreferrer noopener" target="_blank">Announcing Microsoft’s open automation framework to red team generative AI Systems | Microsoft Security Blog</a></li>
</ul>
</li>
</ol>
<p><strong>AI Development Considerations:</strong></p>
<p> </p>
<ol>
<li><span style="font-size: inherit;"><strong>AI Assessment from Microsoft</strong></span>
<ul>
<li><a id="menuran9" class="fui-Link ___1rxvrpe f2hkw1w f3rmtva f1ewtqcl fyind8e f1k6fduh f1w7gpdv fk6fouc fjoy568 figsok6 f1hu3pq6 f11qmguv f19f4twv f1tyq0we f1g0x7ka fhxju0i f1qch9an f1cnd47f fqv5qza f1vmzxwi f1o700av f13mvf36 f1cmlufx f9n3di6 f1ids18y f1tx3yz7 f1deo86v f1eh06m1 f1iescvh fhgqx19 f1olyrje f1p93eir f1nev41a f1h8hb77 f1lqvz6u f10aw75t fsle3fq f17ae5zn" title="https://www.bing.com/aclick?ld=e8i0_yaepp7cxsmf7oo_mmozvucuzunxt-cgfmpjjfwk-spm9wjtvpy_zdze-djrpcyjorva9juporxjjx3p9seckhvszqoc7jyz7_xvwdr7cmvbupaevqcgtielq8gzxqkeqnyfuwnq796d5mcct0n4jbfy6vbjbcluk2lahf0c3hiiglcgdrm5fvky8ysnufvmmsl7kw8aezanh-xdmgrmbv63ly0alxpwqy5ahf4vzhlizmighbwlc6jc7obrggjacurw5liyclsneoocjbj9fggq8&u=ahr0chmlm2elmmylmmz3d3cubwljcm9zb2z0lmnvbsuyzmvulxvzjtjmywklmmzyzxnwb25zawjszs1hasuzzmvmx2lkjtnkx2tfmzgyzjiynmflnjizmthmogeymda0ngq2nwm1mtdhn2rfa18lmjzpq0lejtnkquley21tmw8xznp5nwlfu0vnx19rxzm4mmyymjzhztyymze4zjhhmjawndrknjvjnte3ytdkx2tfjti2bxnjbgtpzcuzzdm4mmyymjzhztyymze4zjhhmjawndrknjvjnte3ytdk&rlid=382f226ae62318f8a20044d65c517a7d" href="https://www.bing.com/aclick?ld=e8i0_Yaepp7CXsMF7oO_MmOzVUCUzUnXT-CGfMPJjFwK-sPm9wjtVPY_Zdze-dJrPCYJORVa9JuPorxjjX3P9seCKhVszqoc7jyZ7_xvWDr7cMVBUpAeVQcgtIelq8gzXQKeqNYFUwnQ796d5mcct0n4jbfY6vbJbcLUK2lahf0C3hiiglCGdRM5fVkY8ySnufVMMsl7KW8AEZAnH-XdMgrMBv63LY0AlXpWQY5AhF4VZHLIZmighBwLc6jC7ObRGGJacURW5lIYCLSneooCjBJ9fggq8&u=aHR0cHMlM2ElMmYlMmZ3d3cubWljcm9zb2Z0LmNvbSUyZmVuLXVzJTJmYWklMmZyZXNwb25zaWJsZS1haSUzZmVmX2lkJTNkX2tfMzgyZjIyNmFlNjIzMThmOGEyMDA0NGQ2NWM1MTdhN2Rfa18lMjZPQ0lEJTNkQUlEY21tMW8xZnp5NWlfU0VNX19rXzM4MmYyMjZhZTYyMzE4ZjhhMjAwNDRkNjVjNTE3YTdkX2tfJTI2bXNjbGtpZCUzZDM4MmYyMjZhZTYyMzE4ZjhhMjAwNDRkNjVjNTE3YTdk&rlid=382f226ae62318f8a20044d65c517a7d" rel="noreferrer noopener" target="_blank"><span style="font-size: inherit;">Conduct an AI assessment using Microsoft’s Responsible AI Impact Assessment Template</span></a></li>
<li><a id="menuranb" class="fui-Link ___1rxvrpe f2hkw1w f3rmtva f1ewtqcl fyind8e f1k6fduh f1w7gpdv fk6fouc fjoy568 figsok6 f1hu3pq6 f11qmguv f19f4twv f1tyq0we f1g0x7ka fhxju0i f1qch9an f1cnd47f fqv5qza f1vmzxwi f1o700av f13mvf36 f1cmlufx f9n3di6 f1ids18y f1tx3yz7 f1deo86v f1eh06m1 f1iescvh fhgqx19 f1olyrje f1p93eir f1nev41a f1h8hb77 f1lqvz6u f10aw75t fsle3fq f17ae5zn" title="https://blogs.microsoft.com/wp-content/uploads/prod/sites/5/2022/06/microsoft-rai-impact-assessment-guide.pdf" href="https://blogs.microsoft.com/wp-content/uploads/prod/sites/5/2022/06/Microsoft-RAI-Impact-Assessment-Guide.pdf" rel="noreferrer noopener" target="_blank"><span style="font-size: inherit;">Responsible AI Impact Assessment Guide</span></a><a id="menurand" class="fui-Link ___1rxvrpe f2hkw1w f3rmtva f1ewtqcl fyind8e f1k6fduh f1w7gpdv fk6fouc fjoy568 figsok6 f1hu3pq6 f11qmguv f19f4twv f1tyq0we f1g0x7ka fhxju0i f1qch9an f1cnd47f fqv5qza f1vmzxwi f1o700av f13mvf36 f1cmlufx f9n3di6 f1ids18y f1tx3yz7 f1deo86v f1eh06m1 f1iescvh fhgqx19 f1olyrje f1p93eir f1nev41a f1h8hb77 f1lqvz6u f10aw75t fsle3fq f17ae5zn" title="https://blogs.microsoft.com/wp-content/uploads/prod/sites/5/2022/06/microsoft-rai-impact-assessment-template.pdf" href="https://blogs.microsoft.com/wp-content/uploads/prod/sites/5/2022/06/Microsoft-RAI-Impact-Assessment-Template.pdf" rel="noreferrer noopener" target="_blank"><span style="font-size: inherit;"> for detailed instructions</span></a></li>
</ul>
</li>
<li><span style="font-size: inherit;"><strong>Microsoft Responsible AI Processes</strong></span>
<ul>
<li><a id="menuranf" class="fui-Link ___1rxvrpe f2hkw1w f3rmtva f1ewtqcl fyind8e f1k6fduh f1w7gpdv fk6fouc fjoy568 figsok6 f1hu3pq6 f11qmguv f19f4twv f1tyq0we f1g0x7ka fhxju0i f1qch9an f1cnd47f fqv5qza f1vmzxwi f1o700av f13mvf36 f1cmlufx f9n3di6 f1ids18y f1tx3yz7 f1deo86v f1eh06m1 f1iescvh fhgqx19 f1olyrje f1p93eir f1nev41a f1h8hb77 f1lqvz6u f10aw75t fsle3fq f17ae5zn" title="https://blogs.microsoft.com/wp-content/uploads/prod/sites/5/2022/06/microsoft-rai-impact-assessment-template.pdf" href="https://blogs.microsoft.com/wp-content/uploads/prod/sites/5/2022/06/Microsoft-RAI-Impact-Assessment-Template.pdf" rel="noreferrer noopener" target="_blank"><span style="font-size: inherit;">Follow Microsoft’s Responsible AI principles: fairness, reliability, safety, privacy, security, inclusiveness, transparency, and accountability</span></a></li>
<li><span style="font-size: inherit;">Utilize tools like the </span><a id="menuranh" class="fui-Link ___1rxvrpe f2hkw1w f3rmtva f1ewtqcl fyind8e f1k6fduh f1w7gpdv fk6fouc fjoy568 figsok6 f1hu3pq6 f11qmguv f19f4twv f1tyq0we f1g0x7ka fhxju0i f1qch9an f1cnd47f fqv5qza f1vmzxwi f1o700av f13mvf36 f1cmlufx f9n3di6 f1ids18y f1tx3yz7 f1deo86v f1eh06m1 f1iescvh fhgqx19 f1olyrje f1p93eir f1nev41a f1h8hb77 f1lqvz6u f10aw75t fsle3fq f17ae5zn" title="https://www.microsoft.com/en-us/ai/responsible-ai" href="https://www.microsoft.com/en-us/ai/responsible-ai" rel="noreferrer noopener" target="_blank"><span style="font-size: inherit;">Responsible AI Dashboard</span></a><a id="menuranj" class="fui-Link ___1rxvrpe f2hkw1w f3rmtva f1ewtqcl fyind8e f1k6fduh f1w7gpdv fk6fouc fjoy568 figsok6 f1hu3pq6 f11qmguv f19f4twv f1tyq0we f1g0x7ka fhxju0i f1qch9an f1cnd47f fqv5qza f1vmzxwi f1o700av f13mvf36 f1cmlufx f9n3di6 f1ids18y f1tx3yz7 f1deo86v f1eh06m1 f1iescvh fhgqx19 f1olyrje f1p93eir f1nev41a f1h8hb77 f1lqvz6u f10aw75t fsle3fq f17ae5zn" title="https://blogs.microsoft.com/wp-content/uploads/prod/sites/5/2022/06/microsoft-rai-impact-assessment-template.pdf" href="https://blogs.microsoft.com/wp-content/uploads/prod/sites/5/2022/06/Microsoft-RAI-Impact-Assessment-Template.pdf" rel="noreferrer noopener" target="_blank"><span style="font-size: inherit;"> for continuous monitoring and improvement</span></a></li>
</ul>
</li>
<li><span style="font-size: inherit;"><strong>Define Use Case and Model Architecture</strong></span>
<ul>
<li><a id="menuranl" class="fui-Link ___1rxvrpe f2hkw1w f3rmtva f1ewtqcl fyind8e f1k6fduh f1w7gpdv fk6fouc fjoy568 figsok6 f1hu3pq6 f11qmguv f19f4twv f1tyq0we f1g0x7ka fhxju0i f1qch9an f1cnd47f fqv5qza f1vmzxwi f1o700av f13mvf36 f1cmlufx f9n3di6 f1ids18y f1tx3yz7 f1deo86v f1eh06m1 f1iescvh fhgqx19 f1olyrje f1p93eir f1nev41a f1h8hb77 f1lqvz6u f10aw75t fsle3fq f17ae5zn" title="https://blogs.microsoft.com/wp-content/uploads/prod/sites/5/2022/06/microsoft-rai-impact-assessment-template.pdf" href="https://blogs.microsoft.com/wp-content/uploads/prod/sites/5/2022/06/Microsoft-RAI-Impact-Assessment-Template.pdf" rel="noreferrer noopener" target="_blank"><span style="font-size: inherit;">Determine the specific use case for your LLM</span></a></li>
<li><a id="menurann" class="fui-Link ___1rxvrpe f2hkw1w f3rmtva f1ewtqcl fyind8e f1k6fduh f1w7gpdv fk6fouc fjoy568 figsok6 f1hu3pq6 f11qmguv f19f4twv f1tyq0we f1g0x7ka fhxju0i f1qch9an f1cnd47f fqv5qza f1vmzxwi f1o700av f13mvf36 f1cmlufx f9n3di6 f1ids18y f1tx3yz7 f1deo86v f1eh06m1 f1iescvh fhgqx19 f1olyrje f1p93eir f1nev41a f1h8hb77 f1lqvz6u f10aw75t fsle3fq f17ae5zn" title="https://blogs.microsoft.com/wp-content/uploads/prod/sites/5/2022/06/microsoft-rai-impact-assessment-template.pdf" href="https://blogs.microsoft.com/wp-content/uploads/prod/sites/5/2022/06/Microsoft-RAI-Impact-Assessment-Template.pdf" rel="noreferrer noopener" target="_blank"><span style="font-size: inherit;">Design the model architecture, focusing on the Transformer architecture</span></a></li>
<li> </li>
</ul>
</li>
<li><span style="font-size: inherit;"><strong>Content Filtering System</strong></span>
<ul>
<li><a id="menuranp" class="fui-Link ___1rxvrpe f2hkw1w f3rmtva f1ewtqcl fyind8e f1k6fduh f1w7gpdv fk6fouc fjoy568 figsok6 f1hu3pq6 f11qmguv f19f4twv f1tyq0we f1g0x7ka fhxju0i f1qch9an f1cnd47f fqv5qza f1vmzxwi f1o700av f13mvf36 f1cmlufx f9n3di6 f1ids18y f1tx3yz7 f1deo86v f1eh06m1 f1iescvh fhgqx19 f1olyrje f1p93eir f1nev41a f1h8hb77 f1lqvz6u f10aw75t fsle3fq f17ae5zn" title="https://learn.microsoft.com/en-us/azure/ai-services/openai/how-to/content-filters" href="https://learn.microsoft.com/en-us/azure/ai-services/openai/how-to/content-filters" rel="noreferrer noopener" target="_blank"><span style="font-size: inherit;">How to use content filters (preview) with Azure OpenAI Service - Azure OpenAI | Microsoft Learn</span></a></li>
<li><span style="font-size: inherit;">Azure OpenAI Service includes a content filtering system that works alongside core models, including DALL-E image generation models. </span><a id="menuranr" class="fui-Link ___1rxvrpe f2hkw1w f3rmtva f1ewtqcl fyind8e f1k6fduh f1w7gpdv fk6fouc fjoy568 figsok6 f1hu3pq6 f11qmguv f19f4twv f1tyq0we f1g0x7ka fhxju0i f1qch9an f1cnd47f fqv5qza f1vmzxwi f1o700av f13mvf36 f1cmlufx f9n3di6 f1ids18y f1tx3yz7 f1deo86v f1eh06m1 f1iescvh fhgqx19 f1olyrje f1p93eir f1nev41a f1h8hb77 f1lqvz6u f10aw75t fsle3fq f17ae5zn" title="https://learn.microsoft.com/en-us/azure/ai-services/openai/concepts/content-filter" href="https://learn.microsoft.com/en-us/azure/ai-services/openai/concepts/content-filter" rel="noreferrer noopener" target="_blank"><span style="font-size: inherit;">This system uses an ensemble of classification models to detect and prevent harmful content in both input prompts and output completions</span></a></li>
<li><a id="menurant" class="fui-Link ___1rxvrpe f2hkw1w f3rmtva f1ewtqcl fyind8e f1k6fduh f1w7gpdv fk6fouc fjoy568 figsok6 f1hu3pq6 f11qmguv f19f4twv f1tyq0we f1g0x7ka fhxju0i f1qch9an f1cnd47f fqv5qza f1vmzxwi f1o700av f13mvf36 f1cmlufx f9n3di6 f1ids18y f1tx3yz7 f1deo86v f1eh06m1 f1iescvh fhgqx19 f1olyrje f1p93eir f1nev41a f1h8hb77 f1lqvz6u f10aw75t fsle3fq f17ae5zn" title="https://learn.microsoft.com/en-us/azure/ai-services/openai/concepts/content-filter" href="https://learn.microsoft.com/en-us/azure/ai-services/openai/concepts/content-filter" rel="noreferrer noopener" target="_blank"><span style="font-size: inherit;">The filtering system covers four main categories: <strong>hate, sexual, violence, and self-harm</strong></span></a></li>
<li><a id="menuranv" class="fui-Link ___1rxvrpe f2hkw1w f3rmtva f1ewtqcl fyind8e f1k6fduh f1w7gpdv fk6fouc fjoy568 figsok6 f1hu3pq6 f11qmguv f19f4twv f1tyq0we f1g0x7ka fhxju0i f1qch9an f1cnd47f fqv5qza f1vmzxwi f1o700av f13mvf36 f1cmlufx f9n3di6 f1ids18y f1tx3yz7 f1deo86v f1eh06m1 f1iescvh fhgqx19 f1olyrje f1p93eir f1nev41a f1h8hb77 f1lqvz6u f10aw75t fsle3fq f17ae5zn" title="https://learn.microsoft.com/en-us/azure/ai-services/openai/concepts/content-filter" href="https://learn.microsoft.com/en-us/azure/ai-services/openai/concepts/content-filter" rel="noreferrer noopener" target="_blank"><span style="font-size: inherit;">Each category is assessed at four severity levels: <strong>safe, low, medium, and high</strong></span></a></li>
<li><span style="font-size: inherit;">Additional classifiers are available for detecting jailbreak risks and known content for text and code. </span><a id="menurao1" class="fui-Link ___1rxvrpe f2hkw1w f3rmtva f1ewtqcl fyind8e f1k6fduh f1w7gpdv fk6fouc fjoy568 figsok6 f1hu3pq6 f11qmguv f19f4twv f1tyq0we f1g0x7ka fhxju0i f1qch9an f1cnd47f fqv5qza f1vmzxwi f1o700av f13mvf36 f1cmlufx f9n3di6 f1ids18y f1tx3yz7 f1deo86v f1eh06m1 f1iescvh fhgqx19 f1olyrje f1p93eir f1nev41a f1h8hb77 f1lqvz6u f10aw75t fsle3fq f17ae5zn" title="https://learn.microsoft.com/en-us/azure/ai-services/openai/how-to/content-filters" href="https://learn.microsoft.com/en-us/azure/ai-services/openai/how-to/content-filters" rel="noreferrer noopener" target="_blank"><span style="font-size: inherit;">JailBreaking Content Filters</span></a></li>
</ul>
</li>
<li><span style="font-size: inherit;"><strong>Red Teaming the LLM</strong></span>
<ul>
<li><a id="menurao3" class="fui-Link ___1rxvrpe f2hkw1w f3rmtva f1ewtqcl fyind8e f1k6fduh f1w7gpdv fk6fouc fjoy568 figsok6 f1hu3pq6 f11qmguv f19f4twv f1tyq0we f1g0x7ka fhxju0i f1qch9an f1cnd47f fqv5qza f1vmzxwi f1o700av f13mvf36 f1cmlufx f9n3di6 f1ids18y f1tx3yz7 f1deo86v f1eh06m1 f1iescvh fhgqx19 f1olyrje f1p93eir f1nev41a f1h8hb77 f1lqvz6u f10aw75t fsle3fq f17ae5zn" title="https://blogs.microsoft.com/wp-content/uploads/prod/sites/5/2022/06/microsoft-rai-impact-assessment-template.pdf" href="https://blogs.microsoft.com/wp-content/uploads/prod/sites/5/2022/06/Microsoft-RAI-Impact-Assessment-Template.pdf" rel="noreferrer noopener" target="_blank"><span style="font-size: inherit;">Plan and conduct red teaming exercises to identify potential vulnerabilities</span></a></li>
<li><a id="menurao5" class="fui-Link ___1rxvrpe f2hkw1w f3rmtva f1ewtqcl fyind8e f1k6fduh f1w7gpdv fk6fouc fjoy568 figsok6 f1hu3pq6 f11qmguv f19f4twv f1tyq0we f1g0x7ka fhxju0i f1qch9an f1cnd47f fqv5qza f1vmzxwi f1o700av f13mvf36 f1cmlufx f9n3di6 f1ids18y f1tx3yz7 f1deo86v f1eh06m1 f1iescvh fhgqx19 f1olyrje f1p93eir f1nev41a f1h8hb77 f1lqvz6u f10aw75t fsle3fq f17ae5zn" title="https://blogs.microsoft.com/wp-content/uploads/prod/sites/5/2022/06/microsoft-rai-impact-assessment-template.pdf" href="https://blogs.microsoft.com/wp-content/uploads/prod/sites/5/2022/06/Microsoft-RAI-Impact-Assessment-Template.pdf" rel="noreferrer noopener" target="_blank"><span style="font-size: inherit;">Use diverse red teamers to simulate adversarial attacks and test the model’s robustness</span></a></li>
<li><a id="menurao7" class="fui-Link ___1rxvrpe f2hkw1w f3rmtva f1ewtqcl fyind8e f1k6fduh f1w7gpdv fk6fouc fjoy568 figsok6 f1hu3pq6 f11qmguv f19f4twv f1tyq0we f1g0x7ka fhxju0i f1qch9an f1cnd47f fqv5qza f1vmzxwi f1o700av f13mvf36 f1cmlufx f9n3di6 f1ids18y f1tx3yz7 f1deo86v f1eh06m1 f1iescvh fhgqx19 f1olyrje f1p93eir f1nev41a f1h8hb77 f1lqvz6u f10aw75t fsle3fq f17ae5zn" title="https://www.microsoft.com/en-us/security/blog/2023/08/07/microsoft-ai-red-team-building-future-of-safer-ai/?msockid=18290294b2206337069616d3b30c629e" href="https://www.microsoft.com/en-us/security/blog/2023/08/07/microsoft-ai-red-team-building-future-of-safer-ai/?msockid=18290294b2206337069616d3b30c629e" rel="noreferrer noopener" target="_blank"><span style="font-size: inherit;">Microsoft AI Red Team building future of safer AI | Microsoft Security Blog</span></a></li>
</ul>
</li>
<li><span style="font-size: inherit;"><strong>Create a Threat Model with OWASP Top 10</strong></span></li>
<ul>
<li><a id="menurao9" class="fui-Link ___1rxvrpe f2hkw1w f3rmtva f1ewtqcl fyind8e f1k6fduh f1w7gpdv fk6fouc fjoy568 figsok6 f1hu3pq6 f11qmguv f19f4twv f1tyq0we f1g0x7ka fhxju0i f1qch9an f1cnd47f fqv5qza f1vmzxwi f1o700av f13mvf36 f1cmlufx f9n3di6 f1ids18y f1tx3yz7 f1deo86v f1eh06m1 f1iescvh fhgqx19 f1olyrje f1p93eir f1nev41a f1h8hb77 f1lqvz6u f10aw75t fsle3fq f17ae5zn" title="https://owasp.org/www-project-top-10-for-large-language-model-applications/assets/pdf/owasp-top-10-for-llms-2023-slides-v1_1.pdf" href="https://owasp.org/www-project-top-10-for-large-language-model-applications/assets/PDF/OWASP-Top-10-for-LLMs-2023-slides-v1_1.pdf" rel="noreferrer noopener" target="_blank">owasp.org/www-project-top-10-for-large-language-model-applications/assets/PDF/OWASP-Top-10-for-LLMs-2023-slides-v1_1.pdf</a></li>
<li><a id="menuraob" class="fui-Link ___1rxvrpe f2hkw1w f3rmtva f1ewtqcl fyind8e f1k6fduh f1w7gpdv fk6fouc fjoy568 figsok6 f1hu3pq6 f11qmguv f19f4twv f1tyq0we f1g0x7ka fhxju0i f1qch9an f1cnd47f fqv5qza f1vmzxwi f1o700av f13mvf36 f1cmlufx f9n3di6 f1ids18y f1tx3yz7 f1deo86v f1eh06m1 f1iescvh fhgqx19 f1olyrje f1p93eir f1nev41a f1h8hb77 f1lqvz6u f10aw75t fsle3fq f17ae5zn" title="https://blogs.microsoft.com/wp-content/uploads/prod/sites/5/2022/06/microsoft-rai-impact-assessment-template.pdf" href="https://blogs.microsoft.com/wp-content/uploads/prod/sites/5/2022/06/Microsoft-RAI-Impact-Assessment-Template.pdf" rel="noreferrer noopener" target="_blank"><span style="font-size: inherit;">Develop a threat model and implement mitigations based on identified risks</span></a></li>
</ul>
</ol>
<p> </p>
<p>Other updates:</p>
<ul>
<li><a href="https://azure.microsoft.com/en-us/updates/v2/Los-Angeles-Azure-Extended-Zones">Los Angeles Azure Extended Zones</a></li>
<li><a href="https://azure.microsoft.com/en-us/updates/v2/Carbon-Optimization">Carbon Optimization</a></li>
<li><a href="https://azure.microsoft.com/en-us/updates/v2/app-config-ref-ga">App Config Ref GA</a></li>
<li><a href="https://azure.microsoft.com/en-us/updates/v2/OS-SKU-in-place-migration-for-AKS">OS SKU In-Place Migration for AKS</a></li>
<li><a href="https://techcommunity.microsoft.com/t5/azure-observability-blog/operator-crd-support-with-azure-monitor-managed-service-for/ba-p/4210644">Operator CRD Support with Azure Monitor Managed Service</a></li>
<li><a href="https://azure.microsoft.com/en-us/updates/v2/Azure-API-Center-Visual-Studio-Code-Extension-Pre-release">Azure API Center Visual Studio Code Extension Pre-release</a></li>
<li><a href="https://azure.microsoft.com/en-us/updates/v2/Azure-API-Management-WordPress-plugin">Azure API Management WordPress Plugin</a></li>
<li><a href="https://azure.microsoft.com/en-us/blog/announcing-a-new-openai-feature-for-developers-on-azure/">Announcing a New OpenAI Feature for Developers on Azure</a></li>
</ul>
Topics
CloudAzureMicrosoftPodcast