--- title: Will Agentic AI Bring Fantasia’s Sorcerer's Apprentice... description: A conversation with Tracy Bannon of MITRE explores the risks of agentic AI, particularly when agents cross boundaries between software ecosystems (e.g.,... canonical: https://daily.dev/posts/will-agentic-ai-bring-fantasia-s-sorcerer-s-apprentice-to-life-a-conversation-with-tracy-bannon-23uswweeu twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Will Agentic AI Bring Fantasia’s Sorcerer's Apprentice to Life?: A Conversation with Tracy Bannon | daily.dev og:description: A conversation with Tracy Bannon of MITRE explores the risks of agentic AI, particularly when agents cross boundaries between software ecosystems (e.g.,... og:url: https://daily.dev/posts/will-agentic-ai-bring-fantasia-s-sorcerer-s-apprentice-to-life-a-conversation-with-tracy-bannon-23uswweeu og:image: https://api.daily.dev/og/posts/23uswWEEU.png og:image:alt: Will Agentic AI Bring Fantasia’s Sorcerer's Apprentice to Life?: A Conversation with Tracy Bannon 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. # Will Agentic AI Bring Fantasia’s Sorcerer's Apprentice to Life?: A Conversation with Tracy Bannon **[InfoQ](https://daily.dev/sources/infoq)** · 46 min read · 0 upvotes · 0 comments ## Summary A conversation with Tracy Bannon of MITRE explores the risks of agentic AI, particularly when agents cross boundaries between software ecosystems (e.g., browser extensions with authenticated sessions). Bannon discusses how AI-driven vulnerability discovery (referencing Anthropic's 'Mythos'/Project Glasswing) is an evolution of existing threats rather than something new, since LLMs like ChatGPT-4 and Claude have already found zero-day vulnerabilities. The discussion covers the unsolved challenge of defining a software development lifecycle for non-deterministic agents, including how to test agents whose behavior drifts ('tunnel vision') during a session, and how to reset test environments. It also addresses growing cognitive overload for developers who must review massive volumes of AI-generated code, and concerns about eroding human skills and critical thinking as reliance on AI assistants grows. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### What security risk comes from using browser extension AI agents like Claude's after logging into a website? Logging into a site and then authorizing a browser extension agent to act creates a security gap because the agent inherits the authenticated session and can take actions, including deletions or transactions, without further verification. One real example involved an agent nearly deleting an Airbnb reservation during an authorized session. The risk grows when agents cross ecosystem boundaries, such as moving from one platform's login into another. _Track emerging guidance on agent authorization risks as browser-based AI tools become part of daily workflows via daily.dev._ ### Why is it hard to test AI agents that use large language models compared to traditional software testing? Agents built on language models are non-deterministic, and repeated conversations within the same context can develop 'tunnel vision,' producing progressively narrower or more biased outputs. Unlike resetting a database between test runs, there is no established equivalent method for resetting an agent's state, so running the same test five times may yield inconsistent results, making repeatability and auditability difficult to guarantee. _Developers building agentic pipelines can follow evolving testing practices for non-deterministic systems on daily.dev._ ### Is Anthropic's Mythos vulnerability-finding AI system a fundamentally new security threat? No, finding and exploiting vulnerabilities with AI models is not new; ChatGPT-4 and Claude models had already demonstrated this capability, and the same flagship vulnerabilities Mythos found were reportedly replicated by smaller, cheaper models. Mythos operates under a controlled-access consortium initially including AWS, Apple, Cisco, Google, and Microsoft, but its ability to convert findings into stealthy exploits is considered weak, with patching remaining the real bottleneck. _Security teams weighing AI-driven vulnerability discovery tools can follow analysis of systems like Mythos on daily.dev._ ## Similar posts on daily.dev - [Fragments: September 29](https://daily.dev/posts/fragments-september-29-hbei69o5x) · Martin Fowler · 0 upvotes · 0 comments - [My fireside chat about agentic engineering at the Pragmatic Summit](https://daily.dev/posts/my-fireside-chat-about-agentic-engineering-at-the-pragmatic-summit-ob1xorish) · Simon Willison · 3 upvotes · 0 comments - [Google begins putting the guardrails on agentic AI](https://daily.dev/posts/google-begins-putting-the-guardrails-on-agentic-ai-gqgdwxx0s) · InfoWorld · 0 upvotes · 0 comments - [Agentic AI in Software Development: From Coding to Orchestration](https://daily.dev/posts/agentic-ai-in-software-development-from-coding-to-orchestration-bl6akhr3q) · Syncfusion · 1 upvotes · 0 comments - [Agents Over Bubbles](https://daily.dev/posts/agents-over-bubbles-n7fscaxsf) · Justin Searls · 1 upvotes · 0 comments --- Tags: [#claude](https://daily.dev/tags/claude), [#ai-security](https://daily.dev/tags/ai-security), [#agentic-ai](https://daily.dev/tags/agentic-ai) [View this post on daily.dev](https://daily.dev/posts/will-agentic-ai-bring-fantasia-s-sorcerer-s-apprentice-to-life-a-conversation-with-tracy-bannon-23uswweeu) ```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":"Will Agentic AI Bring Fantasia’s Sorcerer's Apprentice to Life?: A Conversation with Tracy Bannon","url":"https://daily.dev/posts/will-agentic-ai-bring-fantasia-s-sorcerer-s-apprentice-to-life-a-conversation-with-tracy-bannon-23uswweeu","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/will-agentic-ai-bring-fantasia-s-sorcerer-s-apprentice-to-life-a-conversation-with-tracy-bannon-23uswweeu"},"datePublished":"2026-08-17T11:03:46.672Z","dateModified":"2026-09-14T07:30:38.472Z","description":"A conversation with Tracy Bannon of MITRE explores the risks of agentic AI, particularly when agents cross boundaries between software ecosystems (e.g.,...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/b409d91e8d760e3e969d3a1d715cb235?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/b409d91e8d760e3e969d3a1d715cb235?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"InfoQ","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":"InfoQ","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/afc3bced3e1e4b188dd9127017a60e0c","url":"https://daily.dev/sources/infoq"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/will-agentic-ai-bring-fantasia-s-sorcerer-s-apprentice-to-life-a-conversation-with-tracy-bannon-23uswweeu","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"claude,ai-security,agentic-ai","timeRequired":"PT46M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"InfoQ","item":"https://daily.dev/sources/infoq"},{"@type":"ListItem","position":3,"name":"Will Agentic AI Bring Fantasia’s Sorcerer's Apprentice to Life?: A Conversation with Tracy Bannon"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/will-agentic-ai-bring-fantasia-s-sorcerer-s-apprentice-to-life-a-conversation-with-tracy-bannon-23uswweeu#faq","mainEntity":[{"@type":"Question","name":"What security risk comes from using browser extension AI agents like Claude's after logging into a website?","acceptedAnswer":{"@type":"Answer","text":"Logging into a site and then authorizing a browser extension agent to act creates a security gap because the agent inherits the authenticated session and can take actions, including deletions or transactions, without further verification. One real example involved an agent nearly deleting an Airbnb reservation during an authorized session. The risk grows when agents cross ecosystem boundaries, such as moving from one platform's login into another. Track emerging guidance on agent authorization risks as browser-based AI tools become part of daily workflows via daily.dev."}},{"@type":"Question","name":"Why is it hard to test AI agents that use large language models compared to traditional software testing?","acceptedAnswer":{"@type":"Answer","text":"Agents built on language models are non-deterministic, and repeated conversations within the same context can develop 'tunnel vision,' producing progressively narrower or more biased outputs. Unlike resetting a database between test runs, there is no established equivalent method for resetting an agent's state, so running the same test five times may yield inconsistent results, making repeatability and auditability difficult to guarantee. Developers building agentic pipelines can follow evolving testing practices for non-deterministic systems on daily.dev."}},{"@type":"Question","name":"Is Anthropic's Mythos vulnerability-finding AI system a fundamentally new security threat?","acceptedAnswer":{"@type":"Answer","text":"No, finding and exploiting vulnerabilities with AI models is not new; ChatGPT-4 and Claude models had already demonstrated this capability, and the same flagship vulnerabilities Mythos found were reportedly replicated by smaller, cheaper models. Mythos operates under a controlled-access consortium initially including AWS, Apple, Cisco, Google, and Microsoft, but its ability to convert findings into stealthy exploits is considered weak, with patching remaining the real bottleneck. Security teams weighing AI-driven vulnerability discovery tools can follow analysis of systems like Mythos on daily.dev."}}]} ```