--- title: AI Model Risk Intelligence: Context-Aware Risk Scores... description: Snyk's Evo platform introduces a context-aware AI model risk scoring system built on Likelihood × Impact, where Likelihood is derived from Attack Success Rate... canonical: https://daily.dev/posts/ai-model-risk-intelligence-context-aware-risk-scores-for-every-model-you-deploy-juwoe1sg7 twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: AI Model Risk Intelligence: Context-Aware Risk Scores for Every Model You Deploy | daily.dev og:description: Snyk's Evo platform introduces a context-aware AI model risk scoring system built on Likelihood × Impact, where Likelihood is derived from Attack Success Rate... og:url: https://daily.dev/posts/ai-model-risk-intelligence-context-aware-risk-scores-for-every-model-you-deploy-juwoe1sg7 og:image: https://api.daily.dev/og/posts/JuwoE1sg7.png og:image:alt: AI Model Risk Intelligence: Context-Aware Risk Scores for Every Model You Deploy og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # AI Model Risk Intelligence: Context-Aware Risk Scores for Every Model You Deploy **[Snyk](https://daily.dev/sources/snyk)** · 11 min read · 0 upvotes · 0 comments ## Summary Snyk's Evo platform introduces a context-aware AI model risk scoring system built on Likelihood × Impact, where Likelihood is derived from Attack Success Rate (ASR) measured against real adversarial attacks and Impact reflects deployment-specific consequences. Unlike static safety cards, scores are weighted by deployment archetype — coding agent, chatbot, personal assistant — because the same model carries fundamentally different risks depending on how it's used. The system covers both direct prompt attacks and indirect prompt injection targeting agentic systems. Findings are organized in a three-level taxonomy mapped to OWASP LLM Top 10, OWASP Agentic, MITRE ATLAS, and NIST, enabling teams to move from a headline score to specific attacker goals and connect findings directly to policy enforcement. Snyk telemetry from 3,044 organizations shows that for every known model, roughly 2.8 more AI components run unmanaged, making inventory a prerequisite to risk scoring. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How is Snyk's Evo AI model risk score calculated? The risk score combines likelihood and impact: likelihood comes from Attack Success Rate (ASR), the share of real adversarial attacks (extraction prompts, multi-turn escalation, jailbreaks, tree-of-attack strategies) that succeed against a model behind a baseline system-prompt hardening defense, and impact reflects the damage when an attacker's goal lands. The two are multiplied into a single 0-1000 score, lower being better. _Teams comparing model safety before deployment can follow how scoring methodologies like this evolve on daily.dev._ ### What is indirect prompt injection and why is it harder to catch than direct prompts? Indirect prompt injection embeds malicious instructions inside data an AI agent reads, such as a retrieved document, tool result, or code comment, rather than in what a user types directly. The user never sees the injected instruction, but the agent processes it as context, and if the model follows it, sensitive data can leak, tools can be misused, or unauthorized actions taken. Model-only testing that checks only direct prompts misses this attack class entirely. _Developers securing agentic systems can track emerging prompt injection research on daily.dev._ ### How many unmanaged AI components do organizations typically have beyond the models they know about? For every AI model an organization knows about, roughly 2.8 additional AI components, libraries, MCP servers, and SDKs run unmanaged, according to anonymized AI-BOM telemetry from 3,044 organizations in Snyk's State of Agentic Adoption report Volume II. Some teams estimated a complete AI inventory would take 4 to 5 weeks and 10 to 12 stakeholders to compile. _Teams building an AI inventory process can follow related governance findings on daily.dev._ ## Similar posts on daily.dev - [Why AI Changes the Risk Model for Application Security](https://daily.dev/posts/why-ai-changes-the-risk-model-for-application-security-rxzpdmwfj) · Security Boulevard · 1 upvotes · 0 comments - [Beyond the Score: Using AI to Translate CVEs into Real-World Business Risk](https://daily.dev/posts/beyond-the-score-using-ai-to-translate-cves-into-real-world-business-risk-rgzfktq3c) · Rapid7 Cybersecurity Blog · 0 upvotes · 0 comments - [Snyk’s Evo Reaches 60% of New Deal Volume As AI Risks Soar](https://daily.dev/posts/snyk-s-evo-reaches-60-of-new-deal-volume-as-ai-risks-soar-u5h1cexxo) · SD Times · 0 upvotes · 0 comments - [Old AI Security vs Evo: Watch Agentic Security Replace Weeks of Manual Work](https://daily.dev/posts/old-ai-security-vs-evo-watch-agentic-security-replace-weeks-of-manual-work-emwvtnvqt) · Snyk · 1 upvotes · 0 comments --- Tags: [#security](https://daily.dev/tags/security), [#llm](https://daily.dev/tags/llm), [#ai-security](https://daily.dev/tags/ai-security), [#agentic-ai](https://daily.dev/tags/agentic-ai), [#prompt-injection](https://daily.dev/tags/prompt-injection) [View this post on daily.dev](https://daily.dev/posts/ai-model-risk-intelligence-context-aware-risk-scores-for-every-model-you-deploy-juwoe1sg7) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"AI Model Risk Intelligence: Context-Aware Risk Scores for Every Model You Deploy","url":"https://daily.dev/posts/ai-model-risk-intelligence-context-aware-risk-scores-for-every-model-you-deploy-juwoe1sg7","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/ai-model-risk-intelligence-context-aware-risk-scores-for-every-model-you-deploy-juwoe1sg7"},"datePublished":"2026-08-04T12:02:32.374Z","dateModified":"2026-09-14T08:47:58.970Z","description":"Snyk's Evo platform introduces a context-aware AI model risk scoring system built on Likelihood × Impact, where Likelihood is derived from Attack Success Rate...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/6c69f44906945e5e394536b49bf0deb2?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/6c69f44906945e5e394536b49bf0deb2?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Snyk","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Snyk","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/8fb2a7b471c04bac9af29fbeb3ed1cf6","url":"https://daily.dev/sources/snyk"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/ai-model-risk-intelligence-context-aware-risk-scores-for-every-model-you-deploy-juwoe1sg7","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"security,llm,ai-security,agentic-ai,prompt-injection","timeRequired":"PT11M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Snyk","item":"https://daily.dev/sources/snyk"},{"@type":"ListItem","position":3,"name":"AI Model Risk Intelligence: Context-Aware Risk Scores for Every Model You Deploy"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/ai-model-risk-intelligence-context-aware-risk-scores-for-every-model-you-deploy-juwoe1sg7#faq","mainEntity":[{"@type":"Question","name":"How is Snyk's Evo AI model risk score calculated?","acceptedAnswer":{"@type":"Answer","text":"The risk score combines likelihood and impact: likelihood comes from Attack Success Rate (ASR), the share of real adversarial attacks (extraction prompts, multi-turn escalation, jailbreaks, tree-of-attack strategies) that succeed against a model behind a baseline system-prompt hardening defense, and impact reflects the damage when an attacker's goal lands. The two are multiplied into a single 0-1000 score, lower being better. Teams comparing model safety before deployment can follow how scoring methodologies like this evolve on daily.dev."}},{"@type":"Question","name":"What is indirect prompt injection and why is it harder to catch than direct prompts?","acceptedAnswer":{"@type":"Answer","text":"Indirect prompt injection embeds malicious instructions inside data an AI agent reads, such as a retrieved document, tool result, or code comment, rather than in what a user types directly. The user never sees the injected instruction, but the agent processes it as context, and if the model follows it, sensitive data can leak, tools can be misused, or unauthorized actions taken. Model-only testing that checks only direct prompts misses this attack class entirely. Developers securing agentic systems can track emerging prompt injection research on daily.dev."}},{"@type":"Question","name":"How many unmanaged AI components do organizations typically have beyond the models they know about?","acceptedAnswer":{"@type":"Answer","text":"For every AI model an organization knows about, roughly 2.8 additional AI components, libraries, MCP servers, and SDKs run unmanaged, according to anonymized AI-BOM telemetry from 3,044 organizations in Snyk's State of Agentic Adoption report Volume II. Some teams estimated a complete AI inventory would take 4 to 5 weeks and 10 to 12 stakeholders to compile. Teams building an AI inventory process can follow related governance findings on daily.dev."}}]} ```