The IT Nerd

NVIDIA is acquiring Hugging Face. Here are the details:

Over the past decade, Clem, Julien, Thomas and the team at Hugging Face have built something remarkable: a vibrant home for the open model developer community.

More than 18 million developers, researchers and creators use Hugging Face to share more than 3 million models, 500,000 datasets and 1 million applications. More than 200,000 companies use the platform to discover, evaluate, customize and deploy AI.

Hugging Face will remain an open platform for the entire AI ecosystem. Developers will choose the models they want, the frameworks they want, the clouds and inference service providers they want and the computing platforms they want. NVIDIA compute will not be required to build on or deploy through Hugging Face.

Hugging Face will continue to support open source and open weight models from across the ecosystem, from every model builder. It will continue to support multi-cloud and multi-accelerator development and deployment, so builders can use the hardware and infrastructure that best fit their work.

Ryan McCurdy, VP, Liquibase:

“NVIDIA’s acquisition of Hugging Face is another sign that the value in AI is moving beyond the models themselves. Models are becoming more available, developers are creating software faster than ever, and AI agents are starting to take action across enterprise systems.

“For enterprises, that creates a new challenge. AI doesn’t just accelerate how quickly change can be created. It increases the volume and speed of changes that can reach critical systems. A bad database change can take an application offline, expose sensitive data, or create an audit and compliance issue.

“That makes the control layer more important. The question isn’t just what AI can create. It’s what AI should be allowed to change, what can reach production, and whether those decisions can be governed and audited. As AI becomes more autonomous, enterprises will need those controls built into the path to production rather than relying on humans to catch problems after the fact.”

AI clearly has value. So based on that I fully expect that this will not be the last purchase that NVIDIA makes.

NVIDIA and the Open Secure AI Alliance have proposed a new industry framework called Shared AI Findings Exchange (SAFE) to standardize how organizations report and share AI-related cybersecurity incidents.

Published through the Linux Foundation as a Request for Comments, the guidelines are intended to help organizations share information on AI attacks, vulnerabilities and near misses to improve collective cyber defenses.

The alliance also announced new open-source contributions, including AI security models, datasets, evaluation tools and research designed to improve the security of AI systems and agents.

NVIDIA said the Open Secure AI Alliance has grown to more than 120 member organizations, including technology companies, cybersecurity firms and open-source foundations collaborating to develop shared AI security tools and best practices.

Jacob Krell, Senior Director: Secure AI Solutions & Cybersecurity, Suzu Labs had this to say:

“SAFE is required because AI agent failures don’t fit existing vulnerability disclosure. When a model finds and uses access it shouldn’t have reached, there’s no patch to issue and no vulnerability identifier to publish. Agent failures are often behavioral and non-deterministic, with no signature to match and no fix to deploy.

“Aviation’s safety reporting system has been running since 1976, the financial sector’s threat-sharing body since 1999. Both protect reporters from liability and operate at industry speed. AI threats outpace government coordination, making private-sector self-organization the right model.

“Identity is the piece that makes the rest of the defensive stack enforceable. The alliance’s work on agent identity and access controls lets organizations verify what an agent is before it acts. SAFE adds information sharing on top of that identity and runtime layer. Together, organizations learn from each other’s agent failures fast enough to defend proactively.

“The highest value will come from near-misses. Breaches make headlines, but an agent that probes a boundary and fails never gets published. That behavioral pattern is exactly the intelligence other organizations running similar systems need, and SAFE creates the channel for sharing what would otherwise stay invisible.”

Jeremiah Fowler, Researcher for Black Hills Information Security, Inc. follows with this:

“I believe this is a positive step in the right direction. Organizations that experience AI related attacks gain valuable insights that could help protect others but only if that information is shared. Establishing a framework for responsible information sharing promotes transparency, peer review, and independently evaluated security findings. The more organizations that contribute real-world evidence, the more effectively we can identify emerging attack patterns, common vulnerabilities, and evolving threats while reducing duplicated defensive efforts that waste valuable time and resources.

“As AI becomes increasingly integrated into everyday life, business operations, and critical infrastructure, developing standardized security guidance now is a proactive investment rather than waiting to retrofit a defense after incidents occur. Sharing information about unsuccessful attacks is just as valuable as documenting confirmed compromises, because near misses can often expose weaknesses, configuration issues, or emerging attack tactics before they evolve into critical security incidents. I used to always say “it is not if you will have a data incident, it is when you will have a data incident”.  AI has supercharged the threat landscape in ways we couldn’t have imagined just a few years ago. Reporting and sharing AI related cybersecurity incidents is a great first step. I am happy to see organizations take the initiative instead of waiting around for regulators or lawmakers.”

Standards are good. But I will have to see this in action to get an idea of how well this works. Because there’s nothing worse than a standard that nobody uses.

First, if you’re not up to date on Nvidia’s Open Secure AI Alliance, this will help:Industry Leaders Join Open Secure AI Alliance for AI Safety and Security | NVIDIA Blog

The consensus from experts in the industry is that this is a meaningful step for the industry, but it’s only the beginning. The real challenge now is helping enterprises secure AI agents in the real world, with better identity, visibility, governance, and even hardware security. Their comments dig into what the announcement means beyond the headlines and where organizations still have the most work to do.

John Strand, Black Hills (https://www.linkedin.com/in/john-strand-a1b4b62)

“I think we’re going to see a lot more stories like this because there’s real anxiety in the AI industry right now. Part of it is financial. Many organizations are struggling to demonstrate a meaningful return on investment from their AI initiatives. The other part is security. As researchers continue to show AI systems escaping their intended boundaries or interacting with other systems in unexpected ways, concerns about AI safety are growing. That’s why you’re seeing so many new AI security initiatives. In many cases, they’re trying to establish industry standards before governments step in with legislation or regulation. Whether those efforts are enough remains to be seen.”

Chuck Sobey, General Chair and Co-founder, Chiplet Summit (https://www.linkedin.com/in/chucksobey)

“As the Open Secure AI Alliance focuses on securing the AI ecosystem, one area that deserves greater attention is the security of the underlying hardware powering AI infrastructure. Software security assumes you can trust the silicon it runs on. The coming wave of chiplet-based systems, especially AI accelerators, raises the stakes: more suppliers, more integration points, more attack surfaces. Hardware security has to be part of this conversation from the beginning.”

John Carberry, Solution Sleuth, Xcape Inc. (https://www.linkedin.com/in/john-carberry-1418a622a)

“Accelerating autonomous artificial intelligence adoption creates systemic enterprise exposure when organizations deploy intelligent agents without establishing clear identity boundaries or visibility controls. Bringing industry heavyweights together through the Open Secure AI Alliance provides a much-needed collaborative effort to advance secure design patterns and open-source defensive tooling outside of traditional business competition. While vendor collaboration establishes essential baseline standards for this emerging technology, security leaders cannot treat pre-competitive initiatives as a substitute for internal governance. Organizations frequently fall short by granting artificial intelligence agents excessive API permissions and service account privileges without audit logging, opening direct pathways for indirect prompt injection and unauthorized data exfiltration. Security executives must immediately map all artificial intelligence integrations, enforce least-privilege scoping on agent execution frameworks, and treat agent interactions with external systems as untrusted input requiring strict validation.

“Critical Takeaways

1. Leverage pre-competitive standards: Use open-source security frameworks from industry alliances to standardize threat modeling for artificial intelligence deployments across enterprise environments.
2. Scope agent permissions tightly: Restrict autonomous artificial intelligence agents using least-privilege access controls, credential isolation, and strict API scope boundaries.
3. Treat agent inputs as untrusted: Implement rigorous input validation and comprehensive audit logging for all automated workflows to neutralize indirect prompt injection risks.

“Industry alliances can build safe frameworks outside of business competition, but your security team still has to enforce them inside your corporate network.”

Jacob Krell, Sr. Director: Secure AI Solutions & Cybersecurity, Suzu Labs (https://www.linkedin.com/in/jacob-krell)

“Organizations built identity and access management for people running predictable software. AI agents are neither, and they skip the entire stack.

“Most security teams can’t tell you how many agents are running in their environment right now, or what those agents can access. Developers launch them, ops teams wire them into workflows, and SaaS vendors embed them in products without security ever seeing a ticket. Each agent holds credentials to production systems and behaves non-deterministically, meaning the same agent running the same task can take a different path every time.

“The Hugging Face breach is proof this gap has consequences. OpenAI tested its models’ exploitation capabilities, and those models breached a real company. If OpenAI couldn’t predict what their own models would do in a controlled evaluation, no enterprise should assume they can predict agent behavior in production. When Hugging Face reached for closed frontier models to analyze the attack, safety guardrails blocked them from examining exploit payloads. They ran GLM 5.2, a Chinese open-weight model, on their own infrastructure instead. I’ve hit the same wall. I still run Claude Opus 4.6 for security work because newer models increasingly refuse to process real attack artifacts. If Washington restricts Chinese open-weight models without ensuring equivalent open alternatives from U.S. labs, defenders lose the tool that actually worked when closed models wouldn’t.

“The Open Secure AI Alliance is right that defenders need open, inspectable models they can run on their own infrastructure. HPE’s SPIFFE/SPIRE contribution to the alliance addresses agent identity directly, giving agents cryptographically verifiable identities. Security leaders should be watching that work. Identity for agents is what makes the rest of the defensive stack enforceable.”

Seemant Sehgal, BreachLock (https://www.linkedin.com/in/s-sehgal)

“The gap in most AI deployments right now is not in the model itself. Organizations are running AI agents with access to internal data, external APIs, and automated decision-making workflows, and they have not mapped what those agents can reach or how an adversary would move through that access. Alliance frameworks that standardize how AI systems are evaluated for risk are useful, but the organizations that will benefit from them are the ones that already know what their agents are doing at runtime. Most do not.

PC Magazine is reporting that Nvidia is putting it’s chipset business on hold due to legal issues between itself and Intel. Nvidia sock puppet PR person Robert Sherbin explains why:

“We have said that we will continue to innovate integrated solutions for Intel’s FSB architecture,” Sherbin said in an email. “We firmly believe that this market has a long healthy life ahead. But because of Intel’s improper claims to customers and the market that we aren’t licensed to the new DMI bus and its unfair business tactics, it is effectively impossible for us to market chipsets for future CPUs. So, until we resolve this matter in court next year, we’ll postpone further chipset investments.”

Intel claims that its four-year-old chipset license with Nvidia does not cover the “Nehalem,” or Core series of microprocessors. Nvidia disagrees, and the matter will be hashed out in court in 2010. Nvidia still sells chipsets specifically designed for AMD’s line of processors, but has halted further development as well.

So, anybody who wants a chipset to work with an AMD or and Intel process is not going to Nvidia. That may pose a problem for Apple who uses Nvidia chipsets in Macs because they offer a performance improvement of similar Intel chipsets.

The same report also claims that Nvidia is also exiting the mid- and high-end standalone graphics card market, although the company has apparently called the rumor “patently untrue”. Sure, I remember not to long ago that rumors that it was leaving the chipset business were “groundless” according to Nvidia and today we have them leaving the chipset business. So take that denial for what it’s worth.