As the annual Black Hat conference opens in Las Vegas, the Open Secure AI Alliance — now comprising more than 120 organizations — is advancing new guidelines and open source technologies aimed at strengthening the cybersecurity of agentic AI systems, AI systems that can autonomously pursue goals, select and use tools, and interact with external applications and data.

The Linux Foundation has released a Request for Comments on the Shared AI Findings Exchange (SAFE), a proposed framework designed to transform agentic cybersecurity incidents into shared defensive intelligence for the broader ecosystem. The SAFE guidelines are being developed by an Open Secure AI Alliance working group, with NVIDIA, Cisco, CrowdStrike, Hugging Face, and Red Hat among the members collaborating with the Linux Foundation on the initial proposal. The effort brings together direct competitors — chipmakers, cloud providers, cybersecurity firms, and AI platform companies — under a shared framework, reflecting the urgency of establishing common defensive practices as agentic AI deployments move from pilot projects into production environments.

Under the SAFE guidelines, the alliance proposes to confidentially collect and analyze AI incidents and near misses, notify affected parties, identify recurring control failures, and publish evidence-based operating recommendations intended to reduce systemic risk.

The broader context is one of escalating challenge: every major technology shift introduces new attack surfaces, and defenders must operate at what the alliance describes as "agent speed" to safeguard infrastructure and intellectual property. Open sharing of threat intelligence within trusted ecosystems is positioned as a force multiplier for collective defense.

Alliance Members Expand a Full-Stack Security Toolkit

The SAFE framework complements a growing body of technology contributions from Open Secure AI Alliance members, who are building and sharing open, inspectable tools across the full AI security stack — from identity controls and harnesses to guardrails, logs, and evaluation systems. The emphasis on open source reflects a core premise of the alliance: that security tools which can be independently inspected, audited, and improved by the community are better suited to building trust in systems that can take autonomous action.

NVIDIA's Open Cybersecurity Software and Models

NVIDIA's contributions span the entire stack. The NVIDIA Labs Object-Oriented Agent (NOOA) research harness, available on GitHub, is designed to make agent behavior easier to test, trace, audit, and govern. The NVIDIA OpenShell runtime restricts what an agent can access and execute, enforcing security and privacy controls at the agent level.

NVIDIA's open model families — Nemotron for agentic AI, Cosmos for physical AI, Isaac GR00T for robotics, BioNeMo for healthcare and life sciences, and Alpamayo, described as the world's largest model for autonomous vehicles licensed for commercial use — are released with open weights, datasets, and training techniques.

NVIDIA's open source verified agent skills extend governance to the capability layer. Each skill includes portable instruction sets that are cataloged, scanned for risks such as prompt injection and tool poisoning, cryptographically signed, and documented with a skill card — enabling defenders to verify exactly what a skill does, its origin, and whether it has been modified after publication.

NeMo Guardrails, NeMo Anonymizer, and NeMo Safe Synthesizer help enforce safety policies, protect sensitive data, and generate privacy-safe synthetic data. Garak, NVIDIA's open source LLM vulnerability scanner, allows security teams to test models for data leaks, prompt injections, and jailbreak scenarios prior to deployment.

Identity and Permissions — Who Gets to Act

Securing agent identity is foundational to the broader security stack.

Okta is developing reference implementations for agent identity and access, demonstrating how Cross App Access (XAA), an open protocol, enables AI agents in OpenShell sandbox environments to securely connect to enterprise applications.

Palo Alto Networks has contributed open source tools from Idira, its next-generation identity security platform, including Agent Guard and Agent Watch. These tools help developers apply identity security best practices and safeguards such as securely retrieving secrets for agentic workflows.

A new open source project founded by Red Hat, called asago, maps an organization's custom governance requirements — referencing frameworks such as NIST, OWASP, and the EU AI Act — directly to runtime agent permissions, maintaining a single audit trail from policy clause to live control. As regulators worldwide tighten AI governance requirements, tools that translate policy mandates into enforceable technical controls are becoming increasingly critical for organizations seeking to demonstrate compliance.

Harnesses and Tooling — Orchestrating Security Work

AI agents operate within ecosystems of open and closed models, harnesses, tools, and runtimes. If a model functions as the agent's brain, the harness serves as the body that executes actions through tools, orchestrating how agents are deployed, coordinated, and constrained.

Amazon, one of the newest members of the Open Secure AI Alliance, contributes Strands Agents — an open source toolkit for building AI agents that is open at every layer, giving developers full visibility into agent behavior and the ability to evaluate agentic systems in production. Amazon also contributes Cedar, an open source authorization language that enforces deterministic, verifiable boundaries on agent-permitted actions.

Capital One has open sourced VulnHunter for agentic AI code security. Cloudflare is offering its Vulnerability Discovery Harness as an open source skill for agent systems.

The Microsoft AI Red Team has open sourced several tools and harnesses. PyRIT — the Python Risk Identification toolkit — enables automated red teaming with built-in memory, supporting common targets and custom endpoints. RAMPART and Clarity turn red-team findings and real-world incidents into repeatable tests, and help teams question design assumptions before code is written. Microsoft has also open sourced Assert, which converts natural-language requirements and expected AI safety behaviors into executable evaluations.

Atlas, developed by Wiz, is an autonomous vulnerability research engine that orchestrates specialized AI agents to discover and validate security flaws across code and open source packages.

Visa, which has also joined the Open Secure AI Alliance, has contributed its open sourced Visa Vulnerability Agentic Harness to help teams identify issues and support remediation and validation.

Models — Intelligence Built for AI Safety and Defense

Specialized security and safety models are purpose-built for defense tasks: trained to understand code, locate vulnerabilities, and reason about threats at scale. They can support agentic workflows as systems of models, with both open and closed models working in concert.

Cisco DefenseClaw is an open source agentic governance layer that operates on top of NVIDIA OpenShell, providing automated runtime security when scaling agentic workforces. Cisco has also released two Antares security small language models designed to pinpoint known vulnerabilities within codebases, along with Project CodeGuard to embed secure-by-default practices into AI coding workflows.

CrowdStrike is fine-tuning the NVIDIA Nemotron Nano model for cyber defense. Internal testing achieved 96% accuracy in generating investigation queries within Falcon LogScale, providing a natural-language interface that enhances agent investigative efficiency. CrowdStrike has also published research showing that a specialized NVIDIA Nemotron Nano reasoning model outperforms much larger models on Security Operations Center detection triage while introducing calibrated logit-based confidence to enable measurable, tunable, and auditable autonomous security decisions.

Beyond detection, defenders need to understand why agents acted, whether systems behave safely, and how attacks are evolving. Akamai contributes insights from its State of the Internet reports and Security Intelligence Group research, drawing on real-world data to illuminate AI-era threats.

Cognition has released a trustworthiness evaluation measuring alignment and security risks of open source-derived models, demonstrating that these risks can be mitigated through post-training.

Numbat, developed by Perplexity, is an open source agent security suite for client endpoints. It detects, investigates, and prevents agent activity across macOS, Linux, and Windows, providing defenders with a structured record of agent actions.

Uber has open sourced key components of ADR (Agentic AI Detection and Response), a production system that reconstructs the full causal chain of AI agent activity — from prompt to reasoning, tool calls, and outcomes. ADR currently supports more than 200,000 agent sessions per day across 30,000 endpoints, using a two-tier analysis approach that combines efficient detection with deeper investigation for high-confidence threats.

Availability and Resilience — Rapid Recovery When Moments Count

Agent systems must remain dependable under disruption, contain failures, and recover safely without losing critical state or exposing the broader environment.

LangChain is adding resilience capabilities to its open source frameworks — Deep Agents, LangGraph, and LangChain — enabling agents to retry interrupted work, follow safe recovery paths, resume from saved states, and automatically fall back to alternative models when primary models fail.

Veeam helps organizations keep AI-related data and infrastructure resilient and recoverable through technologies including Kanister, its open source framework for data protection on Kubernetes. Kanister helps teams protect and recover AI workloads, vector databases, and data to a verified known-good state.

Additional contributions are expected as the alliance continues to grow. When members publish reusable mitigations, defenders across the ecosystem can inspect, adapt, and improve them, allowing security practices to evolve alongside advances in AI.

Members of the Open Secure AI Alliance are scheduled to gather at Black Hat on Tuesday, Aug. 4, at 5:15 p.m. PT for a group photo outside the Main Stage, Business Hall at the Mandalay Bay Convention Center. Further information about the alliance is available at the Open Secure AI Alliance page.