Open Secure AI Alliance Unveils SAFE Guidelines One Week After Launch

Nvidia doesn't mess around: A week after open AI industry group formed, it's already showing progress

The week-old Open Secure AI Alliance (OSAA), an industry group spearheaded by Nvidia that has already grown to over 120 companies, has developed a cutely named working group, the Shared AI Findings Exchange, or SAFE. The group is already presenting proposals for open comment, and The Linux Foundation, a member of the group, is managing the proposals. The group developed them while members gathered at the nexus of the cybersecurity world, the Black Hat conference, taking place this week in Las Vegas. The guidelines are nothing terribly earth shattering for now. The proposals cover areas like how to confidentially report AI cybersecurity incidents, alert those affected, and then do blame-free analysis so all can learn from them. At the same time, members of the OSAA are also contributing and cataloging bits and pieces of their open source technology that might be useful. This may, as these types of industry organizations go, eventually coalesce into an open source means for an enterprise to secure their AI agents, or defend against rogue AI attackers, such as the OpenAI model that infiltrated Hugging Face. (Hugging Face is also a member of this group.) For instance, Nvidia has noted that it offers an entire family of open models, as well as an open source LLM vulnerability scanner called Garak; Okta is working on agent identity tech; Red Hat is working on agent governance; Amazon has contributed both an open agent building tool, Strands Agents, and an authorization language Cedar. And there are many more examples. The group now includes a host of big names including Adobe, BlackRock, Cisco, Intel, Microsoft, and Visa, but there are some notable absences, like Anthropic, OpenAI and Google. Interestingly, both OpenAI and Google signed the original open letter that spawned this group. The letter, published last week, urged the White House to support open source AI efforts, not squash them. It was championed by Nvidia and signed by over 200 tech companies. While Anthropic's cold shoulder to the letter and the industry group to date is not a surprise, both OpenAI and Google have released open weight models of their own. Google is generally known as a big supporter of open source, too. We'll see if they join as this group builds momentum. Meanwhile, the group is operating at AI speeds. It's only been a couple of weeks since news broke that the Trump Administration was considering banning Chinese open weight models, causing the industry consternation that resulted in the open letter. Whatever ultimately happens with Chinese open weight models in the U.S., the fast-action by this heavyweight group appears to be a good thing for the U.S. open AI ecosystem Some in the ecosystem, like the co-founder and chief technology officer of U.S. open weight AI lab Arcee, say that's ultimately the way to best any threat -- real or imagined -- that Chinese AI labs pose to the United States. "Openness may be one of the most important paths to AI safety and security," this industry group wrote in their letter. Looks like they are ready to immediately put their effort -- and their tech -- where their mouths are.

AI Leaders Propose SAFE Guidelines for Cybersecurity Transparency

Linux Foundation publishing Request for Comments on Shared AI Findings Exchange (SAFE) guidelines to transform agentic cybersecurity incidents into better protection. Members of the Open Secure AI Alliance -- now more than 120 organizations strong -- are developing new guidelines to strengthen agentic AI cybersecurity as the annual Black Hat conference begins in Las Vegas today. The Linux Foundation today shared a Request for Comments on Shared AI Findings Exchange (SAFE), a proposed set of guidelines designed to turn agentic cybersecurity incidents into shared protection for the entire ecosystem. The SAFE guidelines are being drafted by an Open Secure AI Alliance working group. NVIDIA, Cisco, CrowdStrike, Hugging Face and Red Hat are among Open Secure AI Alliance members working with the Linux Foundation to contribute to the initial proposal. The SAFE guidelines include proposals to confidentially collect and analyze AI incidents and near misses, inform those impacted, identify recurring control failures and publish evidence-based operating recommendations that reduce systemic risk. Cybersecurity is a race without a finish line. Every major technology shift has created new potential attack surfaces. Defenders must move now at agent speed to respond rapidly to protect infrastructure and intellectual property -- and the best way to do that is together. When trusted ecosystems share threat intelligence openly, collective defense becomes a force multiplier. Open Secure AI Alliance Delivers More Tools for AI Cybersecurity The SAFE framework adds to technology contributions Open Secure AI Alliance members are making as part of a shared commitment to building and sharing open, inspectable tools across the full AI security stack. An AI agent isn't just a model. It's a system -- identity controls, harnesses, guardrails, logs and evaluation -- and securing it requires more than vulnerability scanning. Security has always been strongest in the layers -- and in the community's willingness to share what it knows. The hardest problems get solved when defenders learn from each other, openly and at speed. Full Stack of Open NVIDIA Cybersecurity Software and Models NVIDIA's contributions run the length of the stack, starting with the NVIDIA Labs Object-Oriented Agent (NOOA) research harness, on GitHub -- which makes agent behavior easier to test, trace, audit and govern. The NVIDIA OpenShell runtime restricts what an agent can see, touch and do -- enforcing security and privacy controls at the agent level, so an agent can't reach what it shouldn't. NVIDIA's open model families -- NVIDIA Nemotron for agentic AI, NVIDIA Cosmos for physical AI, NVIDIA Isaac GR00T for robotics, NVIDIA BioNeMo for healthcare and life sciences, and NVIDIA Alpamayo, the world's largest model for autonomous vehicles licensed for commercial use -- ship with open weights, datasets and training techniques. NVIDIA open source verified agent skills extend that trust to the capability layer. Each skill provides portable instruction sets -- cataloged, scanned for risks such as prompt injection and tools poisoning, cryptographically signed and documented with a skill card. Defenders know exactly what an agent skill does, where it came from and whether it was modified after publication. NeMo Guardrails, NeMo Anonymizer and NeMo Safe Synthesizer help enforce safety policies, protect sensitive data and generate privacy-safe synthetic data. And Garak, NVIDIA's open source LLM vulnerability scanner, lets security teams check models for data leaks, prompt injections and jailbreak scenarios before they ship. Alliance Members Expand Tools for Open Ecosystem Development Other members of the Open Secure AI Alliance have also been building across the full defensive stack, spanning identity and permissions, harnesses, runtime guardrails, security AI models, observability and evaluation, data security and privacy, availability and resilience, and more. Some of the latest contributions across different layers of the stack are highlighted below, with more continuing to arrive. Identity and Permissions -- Who Gets to Act You can't secure what you can't identify. Okta is developing reference implementations for agent identity and access, showing how Cross App Access (XAA), an open protocol, enables AI agents operating 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 and agent builders apply identity security best practices and safe guards such as securely retrieving secrets for agentic workflows. A new open source project founded by Red Hat, asago takes an organization's custom governance requirements -- such as those referenced in NIST, OWASP and the EU AI Act -- and maps them directly to what agents are allowed to do at runtime, with a single audit trail from policy clause to live control. Harnesses and Tooling -- How Security Work Gets Orchestrated AI agents are more than a model - they tap into systems of open and closed models, harnesses, tools and runtimes to get work done. If a model is the agent's brain, the harness is the body that takes action by using tools. The harness surrounding the model acts like an orchestrator that determines how agents are deployed, coordinated and constrained. Alliance members are contributing tooling, harnesses and supporting technologies across this emerging layer of the AI security stack. Amazon, which today became 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 what AI agents are permitted to do, giving customers fine-grained, analyzable access controls to help ensure only authorized actions reach enterprise resources. Capital One open sourced VulnHunter for agentic AI code security. Cloudflare is offering its Vulnerability Discovery Harness as an open source skill to add security to agent systems. Microsoft AI Red Team has open sourced several tools and harnesses. PyRIT - Python Risk Identification toolkit enables AI red teamers to run automated red teaming, with built in memory, supporting common targets, as well as custom endpoints. RAMPART turns red-team findings and real-world incidents into repeatable tests that run as software changes. Clarity helps teams question design assumptions and identify potential failures before code is written. Microsoft has also open sourced Assert, which converts natural language requirements and expected AI safety and security behaviors into executable evaluations. Atlas is Wiz's autonomous vulnerability research engine that orchestrates specialized AI agents to discover and validate security flaws across code and open source packages. Visa has also joined the Open Secure AI Alliance, contributing its open sourced Visa Vulnerability Agentic Harness to help teams identify issues, support remediation and validation, quickly and safely. Models -- Intelligence Built for AI Safety and Defense Not every security or safety task calls for a general-purpose model. Specialized security and safety models are purpose-built for defense: trained to understand code, locate vulnerabilities and reason about threats at scale. They can work to support agentic workflows as systems of models, with both open and closed models working together to get the job done efficiently. Cisco DefenseClaw is an open source agentic governance layer that sits on top of NVIDIA OpenShell to provide robust, automated security at the runtime level when scaling agentic workforces. Cisco has also released two of its Antares security small language models to help pinpoint where known vulnerabilities exist within a codebase; and Project CodeGuard to embed secure-by-default practices directly 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, delivering a natural-language interface that boosts agent investigative efficiency. CrowdStrike has also published research demonstrating how 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. Seeing what an agent did is only part of the picture. Defenders also need to understand why it acted, whether the system behaves safely and how attacks are evolving in the real world. Akamai brings insights from its State of the Internet reports and Security Intelligence Group research, drawing on real-world data to illuminate AI-era threats and explain how emerging exploits work so defenders can learn, adapt and respond.Cognition has released a trustworthiness evaluation, which measures alignment and security risks of open source-derived models. The evaluation demonstrates these risks can be mitigated via post-training. Numbat is Perplexity's open source agent security suite for client endpoints. It detects, investigates, and prevents agent activity across macOS, Linux and Windows -- giving defenders a structured record of what agents actually did. Uber 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 -- to help security teams detect threats. Today, ADR 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 a safe recovery path, resume from a saved state instead of starting over and automatically fall back to alternative models when the primary model fails. Veeam helps organizations keep the data and infrastructure behind AI resilient and recoverable with technologies such as Kanister, its open source framework for data protection on Kubernetes. It helps teams protect and recover AI workloads, vector databases, and data to a verified known-good state. More contributions are coming. When members publish reusable mitigations, defenders across the ecosystem can inspect, adapt and improve them, helping security practices evolve as AI advances. Join members of the Open Secure AI Alliance at Black Hat today, Tuesday, Aug. 4, at 5:15pm PT, for a group photo outside the Main Stage, Business Hall at the Mandalay Bay Convention Center. Learn more or share interest in joining the Open Secure AI Alliance.

Open Secure AI Alliance proposes SAFE guidelines as membership tops 120

The Open Secure AI Alliance today proposed a set of guidelines for reporting cybersecurity incidents involving artificial intelligence agents, one week after the group was formed. The proposal is called Shared AI Findings Exchange, or SAFE, and was published as a request for comments by the Linux Foundation. Nvidia Corp., Cisco Systems Inc., CrowdStrike Holdings Inc., Hugging Face Inc. and Red Hat Inc. led the drafting. Comments are being taken on GitHub. SAFE would give organizations a confidential channel for handing over details of AI security incidents, agent misbehavior and operational near misses. The alliance would then analyze what it receives, notify the parties affected and flag control failures that keep recurring across its membership. Recommendations would follow, based on the incident evidence rather than on vendor guidance. The alliance launched on July 27 with roughly two dozen founding members and now counts more than 120 organizations. Adobe Inc. and Cloudflare Inc. were in at the start. BlackRock Inc., Capital One Financial Corp., Intel Corp. and Visa Inc. are also on the roster. Anthropic PBC, OpenAI Group PBC and Google LLC have not joined. Its formation followed OpenAI's disclosure on July 21 that two of its models had escaped a sandbox during an internal cyber capability test and used stolen credentials and zero-day exploits to pull test answers off Hugging Face servers. Anthropic followed on July 31 with a separate account of three models attacking targets during evaluations that a configuration error had left connected to the internet. One of the attacks spread to a real cybersecurity company's infrastructure. Members used the Black Hat conference in Las Vegas to detail what code they are handing over. From Okta Inc. come agent identity implementations built on its Cross App Access protocol. Palo Alto Networks Inc. brought two tools, Agent Guard and Agent Watch. The Cedar authorization language and the Strands Agents toolkit are Amazon.com Inc.'s contributions. Microsoft Corp. added PyRIT, its Python risk identification toolkit, plus three other projects. Red Hat is supplying asago, a project that maps written policy to runtime governance. Uber Technologies Inc. handed over ADR, short for agentic AI detection and response. The production system reconstructs the full causal chain of what an agent did, and Uber runs it across more than 200,000 agent sessions a day. Nvidia's contributions include Garak, an open-source vulnerability scanner for large language models, and OpenShell, a runtime that restricts what an agent can see, touch and do. The chipmaker also put its Labs Object-Oriented Agent research harness on GitHub for testing and auditing. Verified agent skills are being released as well. Each is cryptographically signed, scanned for risks including prompt injection and tool poisoning and shipped with a documented skill card. Frank Dickson, group vice president for the security and trust research practice at International Data Corp., said the alliance carries forward the work of Project Glasswing. The focus has shifted from immediate threats to building secure foundations, he said. On validation, he was less certain. "The whole model is a little bit fraught, because open source is contributed, open source is openly managed...how do you validate that?" Dickson told IT Brew. Code-signing efforts may help, he said, though open-source projects run largely on volunteers.

Nvidia's open-source alliance seeks industry input on AI safety controls

A working group within Nvidia's new initiative on open-source technologies is seeking public input as it develops guidelines on how to share and learn from AI cybersecurity incidents to prevent future risks. The request for comment comes about a week after Nvidia announced the Open Secure AI Alliance, a consortium of more than 120 firms focused of building and sharing open-source AI tools to boost cybersecurity defenses. Concerns about the growing capabilities of AI were stoked last month after OpenAI disclosed two of its agents went rogue, escaped an isolated sandbox and breached the systems of technology startup Hugging Face. The guidelines for the sharing of this information will be presented in a framework, dubbed the Shared AI Findings Exchange, Nvidia announced Tuesday. It will include proposals to "confidentially collect and analyze AI incidents and near misses, inform those impacted, identify recurring control failures and publish evidence-based operating recommendations that reduce systemic risk," the technology firm said. "We think it's important to have an open working group that can look at traces when an agent escapes, to be able to confidentially come up with shared recommendations for the industry, on safety controls that would have helped avoid an agent leaking out of an environment," Justin Boitano, the vice president and general manager of enterprise computing at Nvidia, told The Hill. Boitano explained much of the focus from the public is on models, but an "agent harness" is another critical part of the conversation. A harness refers to the software infrastructure around a large language model or AI agent, managing the tools and memory of the models. "As an industry, if we can look at the traces from the harness -- this is like the flight recorder -- you can understand what the agent attempted to do or where systems might not have been set up correctly to prevent the accident," he said. The working group is looking for comment from all corners of the technology ecosystem, including model developers, infrastructure companies, and "AI builders." The request for comment was published by the non-profit Linux Foundation, and the guidelines will be drafted by Nvidia, Cisco, CrowdStrike, Hugging Face and Red Hat. Unlike private models, open-source models live in the public domain where any person or business can download and customize them for personal use. These systems can be used, modified, examined and shared with anyone, for any purpose. In some instances, a model may not be open source but can have open weights, meaning the ways a model is trained to sift through information and formulate answers are made public. Proponents often consider open models to be more transparent, as the entirety of training data, code and process is publicly available. Others argue open-source or open-weight technologies could be misused. Nvidia has acknowledged the risks of open source include the possibility of misuse for cyberattacks, but argued this does not "disappear" in closed systems. "Cyber defenders need open, frontier agentic systems for self-defense," the company wrote. "When closed AI tools -- unable to distinguish attackers from defenders -- blocked essential forensic analysis, Hugging Face ran the open-weight GLM 5.2 model on its own infrastructure to analyze more than 17,000 actions and contain the intrusion," the company said in a release last week, referring to the OpenAI breach.

Nvidia's Open Source Alliance is Now Taking a 'SAFE' Route to Cybersecurity

Their latest proposal recommends that SAFE operate neutrally so that no single vendor or industry segment controls its findings It has been hardly ten days since Nvidia spearheaded the Open Secure AI Alliance (OSAA) with Microsoft, Intel and Cisco also joining hands to promote open-weight AI models and shift the US narrative around it. Now the group's strength has grown to 120 signatory companies. And they have created a working group called Shared AI Findings Exchange (SAFE) that is presenting proposals for open comment. And in charge of the open comments and managing such proposals is the Linux Foundation, one of the early members formed by the group that had dashed off an open letter on July 24 urging the White House to support open source AI efforts and not squash them. Last evening, the Linux Foundation posted a request for comments on its website on the draft proposal "intended to start an open community discussion around how organizations can confidentially learn from AI security incidents and near misses, transforming those experiences into practical guidance that strengthens security across the broader AI ecosystem." The OSAA reportedly developed the proposal while members were attending the Black Hat conference taking place this week in Las Vegas. While the guidelines do not indicate any major shifts, covering areas like confidentiality on reporting AI cybersecurity incidents, sending alerts to those impacted and follow it up with a "blame-free analysis" so that everyone is wiser. The proposal recommends that "SAFE operate neutrally so that no single vendor or industry segment controls its findings. Its processes are intended to apply equally to both open and proprietary AI systems. As the proposal states, trust is not a security control. Shared evidence and verifiable improvements are how trust is earned." A separate post shared by Nvidia said members of the OSAA are also contributing and cataloguing bits of information about their open source technologies that may render useful in the future and become something of an open source means for enterprises to secure their AI agents or defend themselves against rogue AI attackers resulting from inadvertent means or otherwise. The blog post said "Nvidia, Cisco, CrowdStrike, Hugging Face and Red Hat are among Open Secure AI Alliance members working with the Linux Foundation to contribute to the initial proposal." Nvidia also shared that it was offering an entire stack of open models as well as an open source LLM vulnerability scanned called Garak. Meanwhile Red Hat was working on agent governance while Amazon has contributed both an open agent building tool (Strands Agents) and an authorisation language called Cedar. While OpenAI and Google had signed the original open letter that helped create this group, Anthropic has thus far given it the cold shoulder. Both OpenAI and Google have released open weight models of their own, while Dario Amodei clarified that he wasn't against open-weight models but his concern stemmed from possible security threats from Chinese models. Of course, the question still remains on how President Trump responds to the Chinese open-weight models being vouched by US enterprises. The group that wants him to go slow with his response are clear in their minds. They noted in the letter that "Openness may be one of the most important paths to AI safety and security."