Research, product and essays from Vijil on trustworthy AI agents.https://www.vijil.ai/https://www.vijil.ai/blog/adaptive-red-teaming-in-vijil-diamond/https://www.vijil.ai/blog/adaptive-red-teaming-in-vijil-diamond/Adaptive Red Teaming runs multi-turn adversarial tests built from your agent’s own policies, tools, and personas, inside your VPC. On the DTap benchmark it succeeded on 1.38 times as many attack tasks as the benchmark’s own curated attacks.Thu, 08 Oct 2026 12:00:00 GMTVijilAnnouncementshttps://www.vijil.ai/blog/look-inside-diamond-adaptive-red-teaming/https://www.vijil.ai/blog/look-inside-diamond-adaptive-red-teaming/How DART plans multi-turn attacks from an agent’s own profile, runs them in waves against the live agent, and reflects on each wave to sharpen the next.Tue, 06 Oct 2026 12:00:00 GMTVele TosevskiResearch noteshttps://www.vijil.ai/blog/out-of-the-shadows-know-every-agent-govern-every-agent-trust-every-agent/https://www.vijil.ai/blog/out-of-the-shadows-know-every-agent-govern-every-agent-trust-every-agent/You cannot govern a population of agents you cannot categorize, register and fingerprint. Shadow AI and ungoverned AI are different problems, and discovery is where the harder one begins.Thu, 24 Sep 2026 12:00:00 GMTSteve CoplanEssayshttps://www.vijil.ai/blog/trust-is-the-gate-the-ciso-consensus-catches-up/https://www.vijil.ai/blog/trust-is-the-gate-the-ciso-consensus-catches-up/We argued that what gates agent autonomy is not capability but trust. Mayfield's 2026 survey of 60-plus security leaders says the CISO consensus has caught up.Wed, 23 Sep 2026 12:00:00 GMTSteve CoplanEssayshttps://www.vijil.ai/blog/vijil-fundamental-research---a-new-path-to-scalable-verifiable-ai-governance/https://www.vijil.ai/blog/vijil-fundamental-research---a-new-path-to-scalable-verifiable-ai-governance/How do organizations deploy more agents with greater autonomy when oversight and audit are limited by human review? New Vijil research develops a neuro-symbolic approach that makes agent behavior auditable, oversight scalable, and flags policy violations more reliably than the strongest LLM-as-Judges.Wed, 23 Sep 2026 12:00:00 GMTAbhishek KulkarniResearch noteshttps://www.vijil.ai/blog/staying-local-why-the-on-premise-option-matters-for-agentic-ai-security/https://www.vijil.ai/blog/staying-local-why-the-on-premise-option-matters-for-agentic-ai-security/Most regulated and security-conscious organizations already have a definitive answer to the question of: "where do our data, logs, and security events live”. Adoption of AI technologies introduces new classes of data from inference and reasoning traces, guardrail actions, policy violations, prompt inputs and outputs, behavioral traces, as well as system logs. Where that data ends up is now of some immediate concern to compliance and governance teams.Sun, 30 Aug 2026 12:00:00 GMTSteve CoplanCompliancehttps://www.vijil.ai/blog/how-exactly-does-vijil-handle-agent-identity/https://www.vijil.ai/blog/how-exactly-does-vijil-handle-agent-identity/No passwords, no keys in env vars. An agent's identity is minted from what it is verifiably, and proved every time it exercises an authorized capability.Sun, 16 Aug 2026 12:00:00 GMTVin SharmaHow-tohttps://www.vijil.ai/blog/genetic-sequence/https://www.vijil.ai/blog/genetic-sequence/Three essays from November 2024, republished with their original dates — the genome from which Vijil grew.Sat, 15 Aug 2026 12:00:00 GMTVin SharmaEssayshttps://www.vijil.ai/blog/the-guardrails-gap-open-source-quick-fix-pros-and-cons/https://www.vijil.ai/blog/the-guardrails-gap-open-source-quick-fix-pros-and-cons/Six months ago, "guardrails" were a checkbox on an AI governance slide. Today, for most engineering teams, they're an urgent, unplanned line item — and open-source guardrails have become the fastest way to get one in place.Wed, 05 Aug 2026 12:00:00 GMTSteve CoplanEssayshttps://www.vijil.ai/blog/iso-42001-compliance-the-path-to-operationalizing-agentic-ai/https://www.vijil.ai/blog/iso-42001-compliance-the-path-to-operationalizing-agentic-ai/How can engineering and development teams prove to security, legal, and compliance teams they have done the hard work to assure agents are trusted and they have been assessed for adversarial robustness before deployment, and that a monitoring system is in place to respond when trust degrades and failures happen - whether because of performance, accuracy, drift, and bias issues?Mon, 27 Jul 2026 12:00:00 GMTSteve CoplanCompliancehttps://www.vijil.ai/blog/untangling-the-agentic-ai-governance-bottleneck/https://www.vijil.ai/blog/untangling-the-agentic-ai-governance-bottleneck/Building agents is the easy part. Even building a cross-functional AI governance committee to define policies for deploying these agents is within reach for most organizations. What many organizations are contending with is dealing with the structural, functional and technical challenges to embedding governance across the full lifecycle of an AI agent.Wed, 15 Jul 2026 12:00:00 GMTSteve CoplanCompliancehttps://www.vijil.ai/blog/governance-gap/https://www.vijil.ai/blog/governance-gap/Governance teams can define policies, yet have no reliable mechanism to confirm that an AI system complies — or keeps complying as it changes. Here is what a closed-loop alternative looks like.Mon, 22 Jun 2026 12:00:00 GMTVijilPartnershipshttps://www.vijil.ai/blog/embedding-trust/https://www.vijil.ai/blog/embedding-trust/A geometric signal — semantic isotropy — flags unreliable long-form output with no labels, no fine-tuning, and no access to model internals.Tue, 16 Jun 2026 12:00:00 GMTVijil ResearchResearch noteshttps://www.vijil.ai/blog/how-much-trust-is-enough-for-ai-governance/https://www.vijil.ai/blog/how-much-trust-is-enough-for-ai-governance/When organizations rush agentic AI into production, they keep bumping into the same uncomfortable question: how much trust is enough? In a recent webinar, How Much Trust Is Enough? Mastering Agentic AI Governance , AI security luminary Ken Huang and Vijil founder and CEO Vin Sharma worked through the question from first principles—and arrived at an answer from different perspectives.Sun, 07 Jun 2026 12:00:00 GMTSteve CoplanEssayshttps://www.vijil.ai/blog/trust-is-the-real-bar-for-production-ai-agents/https://www.vijil.ai/blog/trust-is-the-real-bar-for-production-ai-agents/One of the key themes at QCon AI event in Boston this week was Governance: Ensuring security and compliance in non-deterministic systems.Wed, 03 Jun 2026 12:00:00 GMTVijilEssayshttps://www.vijil.ai/blog/from-policy-to-code-using-the-naic-model-to-govern-ai-agents/https://www.vijil.ai/blog/from-policy-to-code-using-the-naic-model-to-govern-ai-agents/Does the rate of change, the broad penetration, and the complexity of managing and securing AI in general, and AI agents in particular, mean that organizations are stuck on the ground - or flying blind - without a viable strategy for operationalizing governance and dealing with regulatory pressures?Sat, 23 May 2026 12:00:00 GMTSteve CoplanCompliancehttps://www.vijil.ai/blog/mapping-the-unknown-open-problems-in-frontier-ai-risk-management/https://www.vijil.ai/blog/mapping-the-unknown-open-problems-in-frontier-ai-risk-management/Frontier AI systems — general-purpose models capable of performing a vast range of tasks — represent a qualitative leap beyond the narrow AI applications that existing risk management standards were built to govern. A research paper from the Oxford Martin AI Governance Initiative, co-authored by over 30 researchers across academia, industry, and policy, sets out to do something overdue: systematically identify and characterize the open problems that stand between where frontier AI risk management is today and where it needs to be in order to support more coordinated and effective progress.Wed, 29 Apr 2026 12:00:00 GMTVijilResearch noteshttps://www.vijil.ai/blog/leading-ai-researcher-tim-g-j-rudner-joins-vijil-as-chief-scientist/https://www.vijil.ai/blog/leading-ai-researcher-tim-g-j-rudner-joins-vijil-as-chief-scientist/Vijil research team focuses on fundamental challenges of building trustworthy AI agents for enterprise useTue, 28 Apr 2026 12:00:00 GMTVijilAnnouncementshttps://www.vijil.ai/blog/crossing-the-agent-pilot-to-production-chasm/https://www.vijil.ai/blog/crossing-the-agent-pilot-to-production-chasm/Agentic AI holds the promise of transforming how work is done, and enterprises function. The challenge is how to transform that promise into tangible reality, where the demonstrable benefits consistently outweigh the known risks.Tue, 31 Mar 2026 12:00:00 GMTSteve CoplanEssayshttps://www.vijil.ai/blog/vijil-launches-platform-enabling-ai-agents-to-adapt-to-attacks-and-failures/https://www.vijil.ai/blog/vijil-launches-platform-enabling-ai-agents-to-adapt-to-attacks-and-failures/New capability allows enterprises to continuously improve agent resilienceWed, 18 Mar 2026 12:00:00 GMTVijilAnnouncementshttps://www.vijil.ai/blog/towards-proactive-compliance-why-partial-preferences-matter/https://www.vijil.ai/blog/towards-proactive-compliance-why-partial-preferences-matter/AI agents are increasingly deployed in high-stakes settings — customer service, healthcare, legal workflows. In these settings, agents engage in multi-turn interactions where decisions unfold over time, and compliance requires satisfying preferences over sequences of actions, not just individual outputs.Mon, 09 Mar 2026 12:00:00 GMTAbhishek KulkarniResearch noteshttps://www.vijil.ai/blog/revisiting-llm-security-scanning-with-garak-in-the-agentic-era/https://www.vijil.ai/blog/revisiting-llm-security-scanning-with-garak-in-the-agentic-era/Recent advances in agentic AI have been driven by rapid innovations in LLM reasoning and inference. This is forcing teams building AI agents to rethink how they evaluate LLMs and their properties.Mon, 02 Mar 2026 12:00:00 GMTLeif Hancox-LiHow-tohttps://www.vijil.ai/blog/2026-the-year-that-businesses-get-ai-agents-they-can-trust/https://www.vijil.ai/blog/2026-the-year-that-businesses-get-ai-agents-they-can-trust/Republished from VM BlogThu, 22 Jan 2026 12:00:00 GMTVin SharmaEssayshttps://www.vijil.ai/blog/vijil-raises-17-million-to-make-ai-agents-resilient-named-a-gartner-cool-vendor/https://www.vijil.ai/blog/vijil-raises-17-million-to-make-ai-agents-resilient-named-a-gartner-cool-vendor/Customers such as SmartRecruiters deploy trusted AI agents 75% faster with VijilMon, 24 Nov 2025 12:00:00 GMTVin SharmaAnnouncementshttps://www.vijil.ai/blog/deploy-vijil-dome-to-defend-ai-agents-on-digitalocean-kubernetes/https://www.vijil.ai/blog/deploy-vijil-dome-to-defend-ai-agents-on-digitalocean-kubernetes/Deploying Vijil Dome to defend your AI agents running on a DigitalOcean Kubernetes cluster is now easierTue, 11 Nov 2025 12:00:00 GMTVarun CherukuriPartnershipshttps://www.vijil.ai/blog/how-can-ai-agents-be-audited-automatically/https://www.vijil.ai/blog/how-can-ai-agents-be-audited-automatically/Phala + Vijil deploy trusted agents in a TEETue, 07 Oct 2025 12:00:00 GMTVele TosevskiPartnershipshttps://www.vijil.ai/blog/trusted-agents-at-scale/https://www.vijil.ai/blog/trusted-agents-at-scale/Groq delivers inference speed at scale; Vijil delivers trust. Together they let you ship an agent that is fast and defended, rather than choosing between the two.Mon, 06 Oct 2025 12:00:00 GMTAnuj TambwekarPartnershipshttps://www.vijil.ai/blog/from-swords-to-plowshares-generating-guardrails-from-evaluations/https://www.vijil.ai/blog/from-swords-to-plowshares-generating-guardrails-from-evaluations/Announcing a new feature of Vijil Evaluate that enables you to automatically generate custom guardrails based on your agent evalSun, 14 Sep 2025 12:00:00 GMTAnuj TambwekarHow-tohttps://www.vijil.ai/blog/exploring-chatbot-mistakes-from-root-causes-to-potential-solutions/https://www.vijil.ai/blog/exploring-chatbot-mistakes-from-root-causes-to-potential-solutions/In this reflection from our Marketing Intern, Lena, we explore why chatbots still make toxic, biased, or baffling mistakes—and how transparency, better training practices, and tools like Vijil’s can help us build AI agents that reflect the best of human values.Mon, 04 Aug 2025 12:00:00 GMTLena KipnisEssayshttps://www.vijil.ai/blog/how-secure-is-your-agent-under-the-vijil-dome/https://www.vijil.ai/blog/how-secure-is-your-agent-under-the-vijil-dome/Discover how to protect your LangChain agents from prompt injection, data leaks, and harmful outputs using Vijil Dome’s guardrails and track improvements with Vijil Evaluate.Tue, 29 Jul 2025 12:00:00 GMTVarun CherukuriHow-tohttps://www.vijil.ai/blog/customizing-guardrails-with-vijil-dome/https://www.vijil.ai/blog/customizing-guardrails-with-vijil-dome/In Part 3 of our Vijil Dome series, learn how to build custom guardrails that enforce domain-specific protections—like PII detection—on top of Dome’s secure, production-ready AI safety infrastructure.Mon, 28 Jul 2025 12:00:00 GMTVarun CherukuriHow-tohttps://www.vijil.ai/blog/defending-ai-agents-with-vijil-dome-langchain/https://www.vijil.ai/blog/defending-ai-agents-with-vijil-dome-langchain/Learn how Vijil Dome integrates seamlessly with LangChain to add powerful, chain-native security guardrails for building robust, secure, and production-ready AI workflows.Sun, 27 Jul 2025 12:00:00 GMTVarun CherukuriHow-tohttps://www.vijil.ai/blog/defending-ai-agents-with-vijil-dome-openai-clients/https://www.vijil.ai/blog/defending-ai-agents-with-vijil-dome-openai-clients/In this post, we introduce Vijil Dome, our open-source guardrails framework designed for real-time, policy-based control of agent behavior. Dome offers low-latency, framework-agnostic enforcement that integrates seamlessly with OpenAI, LangChain, and other popular stacks. Learn how to wrap your AI workflows with trusted perimeter defense, customize policies for your industry, and start building secure, compliant agents—without compromising performance.Wed, 23 Jul 2025 12:00:00 GMTVarun CherukuriHow-tohttps://www.vijil.ai/blog/vijil-dome-securing-the-future-of-ai-agents/https://www.vijil.ai/blog/vijil-dome-securing-the-future-of-ai-agents/At AWS Summit NYC, a shift toward enterprise-ready AI agents became clear. Jamie Mneimneh (Brightmind Partners) and Vin Sharma (Vijil) break down why security must be built into agent infrastructure—not bolted on. They introduce Vijil Dome, a guardrails framework that enforces real-time input/output controls, tool restrictions, and privacy safeguards. Paired with AgentCore, it creates scalable, accountable AI systems ready for production.Mon, 21 Jul 2025 12:00:00 GMTGuest ContributorEssayshttps://www.vijil.ai/blog/protecting-google-adk-agents-with-vijil-dome/https://www.vijil.ai/blog/protecting-google-adk-agents-with-vijil-dome/Building secure AI agents is just the start—defending them is where it counts. Follow our step-by-step guide to integrate Vijil Dome into your ADK agent and see how it boosts protection and trust in real-world scenarios.Tue, 20 May 2025 12:00:00 GMTAnuj TambwekarHow-tohttps://www.vijil.ai/blog/test-the-trustworthiness-of-agents-built-with-google-adk/https://www.vijil.ai/blog/test-the-trustworthiness-of-agents-built-with-google-adk/Google Agent Development Kit (ADK) is a powerful framework for building AI agents. But as every agent developer knows by now, creating an agent is the easy part. Making sure that it runs reliably, securely, and safely is what really tests your mettle. ADK provides an agent evaluations capability to help you test your agents. But you need to define test files, hand-crafting tests to verify that your agent is ready for production. Creating a test harness that is sufficiently comprehensive and diverse is no easy task. This is where Vijil can help.Mon, 12 May 2025 12:00:00 GMTAnuj TambwekarHow-tohttps://www.vijil.ai/blog/defending-deepseek-r1-with-vijil-dome/https://www.vijil.ai/blog/defending-deepseek-r1-with-vijil-dome/While DeepSeek R1 is competitive to OpenAI GPT-o1 on reasoning tasks, it has several failure modes that hamper its use in business-critical applications. Vijil Dome adds a layer of trust around the model to make it viable in commercial products.Mon, 05 May 2025 12:00:00 GMTAnuj TambwekarHow-tohttps://www.vijil.ai/blog/vijil-named-to-the-2025-cb-insights-list-of-the-100-most-innovative-ai-startups/https://www.vijil.ai/blog/vijil-named-to-the-2025-cb-insights-list-of-the-100-most-innovative-ai-startups/Vijil recognized for innovations in trustworthy AI agentsWed, 23 Apr 2025 12:00:00 GMTVijilAnnouncementshttps://www.vijil.ai/blog/vijil-announces-listing-of-vijil-evaluate-on-google-cloud-marketplace/https://www.vijil.ai/blog/vijil-announces-listing-of-vijil-evaluate-on-google-cloud-marketplace/Vijil announced today that Vijil Evaluate, its AI agent testing service, is now available on Google Cloud Marketplace to help developers test agents with rigor and speed so that they can deploy their agents into production sooner.Tue, 01 Apr 2025 12:00:00 GMTVijilAnnouncementshttps://www.vijil.ai/blog/ai-ethics-benchmarks-are-failing/https://www.vijil.ai/blog/ai-ethics-benchmarks-are-failing/Are AI ethics benchmarks actually useful? Many companies rely on abstract, outdated tests that don’t reflect real-world compliance. Discover why Vijil’s AI evaluation applies company-specific ethical standards to ensure trustworthy AI—just like you would with an employee. Test your AI in minutes today!Mon, 17 Mar 2025 12:00:00 GMTVin SharmaEssayshttps://www.vijil.ai/blog/factotum/https://www.vijil.ai/blog/factotum/Today’s AI agents are factotums — competent but loyal to their developers. What we deserve are fiduciary agents with a duty of care.Sat, 30 Nov 2024 12:00:00 GMTVin SharmaEssayshttps://www.vijil.ai/blog/deconstructing-trust/https://www.vijil.ai/blog/deconstructing-trust/We cannot trust AI agents the way we trust machines or people. Perhaps we ought to trust them the way we trust corporations.Fri, 22 Nov 2024 12:00:00 GMTVin SharmaEssayshttps://www.vijil.ai/blog/trust-in-agents/https://www.vijil.ai/blog/trust-in-agents/The first in a series excavating the root of trust — and why we cannot trust AI agents the way we trust machines or people.Thu, 14 Nov 2024 12:00:00 GMTVin SharmaEssayshttps://www.vijil.ai/blog/cheaper-and-faster-mmlu-score/https://www.vijil.ai/blog/cheaper-and-faster-mmlu-score/Run MMLU-Pro benchmarks 20x cheaper and 1000x faster with Vijil Evaluate. Discover how our high-performance evaluation engine and Lite benchmarks deliver 95% accuracy at a fraction of the cost and time. Optimize your LLM testing today!Wed, 11 Sep 2024 12:00:00 GMTAditya KaranEssayshttps://www.vijil.ai/blog/vijil-emerges-from-stealth/https://www.vijil.ai/blog/vijil-emerges-from-stealth/Vijil came out of stealth on July 24 with funding from Mayfield's AIStart seed fund and Gradient Ventures, Google's AI-focused seed fund. We are thrilled to have them as partners backing our venture.Tue, 27 Aug 2024 12:00:00 GMTVijilAnnouncementshttps://www.vijil.ai/blog/evaluating-meta-llama-3-for-stereotyping/https://www.vijil.ai/blog/evaluating-meta-llama-3-for-stereotyping/To help organizations build AI agents that humans can trustFri, 26 Apr 2024 12:00:00 GMTVin SharmaResearch Noteshttps://www.vijil.ai/blog/conversation-with-garak-founder/https://www.vijil.ai/blog/conversation-with-garak-founder/Catch our live conversation with Leon Derczynski, creator of garak, the leading open-source LLM vulnerability scanner. We explore the motivations behind garak, how it works, and how it’s helping teams secure AI systems against real-world threats.Wed, 24 Jan 2024 12:00:00 GMTSubho MajumdarEssays