Open Secure AI Alliance aims to open-source AI security defences

Open Secure AI Alliance aims to open-source AI security defences Skip to content

Nvidia and the Alliance are using that episode as the case for the entire initiative. When a closed model’s safety filters can’t tell an incident responder from an intruder, a defender locked into that one vendor has few options left; they either wait for a fix or lose visibility during the window when speed matters most.

Hugging Face had an alternative it controlled directly, but not every enterprise runs a research lab with spare model weights on hand. Closing that gap is the argument members are making for treating open frontier models as infrastructure rather than a niche option for hobbyist developers.

None of that resolves the case against open models outright. Weights any defender can inspect are also weights an attacker can inspect, strip of safeguards, or fine-tune for misuse, and the Alliance doesn’t dispute that.

The position of the Open Secure AI Alliance is that closed weights don’t remove the risk either, they just put it behind a vendor’s access controls. The group argues that pairing openness with evaluation, monitoring, and quick patching gives defenders more capacity than restricting access does.

Governance below the model layer

An AI agent used for security work is a stack, not a single model. Identity checks, permission scopes, a harness that constrains what the model can call, logging, and evaluation all sit around whatever weights run the reasoning. Members are contributing pieces of that stack rather than competing on a single open-versus-closed model release.

Nvidia’s contribution is the NVIDIA Labs Object-Oriented Agent (NOOA) project, released on GitHub as open-source research code. NOOA targets the harness layer specifically, the software sitting between a model and the actions it’s permitted to take, and it’s built to make agent behaviour easier to trace, test, audit, and govern rather than to improve raw model reasoning. After all, a model can be technically-capable and still produce agents nobody can audit if the harness around it doesn’t log or constrain what it does.

Other members are filling adjacent gaps. HPE is contributing to SPIFFE/SPIRE, an identity framework that issues cryptographic identities to workloads so systems can verify an AI agent is what it claims to be before granting access to enterprise resources.

Hugging Face has handed Safetensors, its format for storing model weights without embedded executable code, to the PyTorch Foundation, addressing a known route for remote code execution in older model file formats.

IBM and Red Hat’s Lightwell project applies digitally-signed patches across the open-source supply chain. Microsoft is contributing MDASH, a scanning harness that runs multiple specialised agents against code to find and prove exploitable bugs rather than simply flag suspicious patterns. xAI has open-sourced its Grok Build coding agent and says it plans to release the weights of its Grok model line to outside researchers.

What regulators do with this claim

The Alliance’s public argument is aimed at policymakers currently drafting AI safety rules: treat open weights, harnesses, and security tooling as assets defenders rely on, not a category to restrict by default.

A blanket rule against open frontier models, the group argues, would concentrate AI defence capacity inside a small number of closed vendors and remove the option Hugging Face used during its own incident.

Whether that argument holds will depend on cases beyond this one. The Alliance’s founding document leans on a single documented incident and a set of contributions, several released only this week without a track record in hostile production environments. NOOA remains a research framework on GitHub rather than a supported product with a maintenance contract behind it.

Members will need more recovery stories like Hugging Face’s before open agent stacks are treated as the default choice rather than a backup option.

See also: Codeberg members vote to reject LLM training and vibe coding

Want to learn more about cybersecurity from industry leaders? Check out Cyber Security & Cloud Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events including the AI & Big Data Expo. Click here for more information.

Developer is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.

To provide the best experiences, we use technologies like cookies to store and/or access device information. Consenting to these technologies will allow us to process data such as browsing behavior or unique IDs on this site. Not consenting or withdrawing consent, may adversely affect certain features and functions.

The technical storage or access is strictly necessary for the legitimate purpose of enabling the use of a specific service explicitly requested by the subscriber or user, or for the sole purpose of carrying out the transmission of a communication over an electronic communications network.

The technical storage or access is necessary for the legitimate purpose of storing preferences that are not requested by the subscriber or user.

The technical storage or access that is used exclusively for statistical purposes. The technical storage or access that is used exclusively for anonymous statistical purposes. Without a subpoena, voluntary compliance on the part of your Internet Service Provider, or additional records from a third party, information stored or retrieved for this purpose alone cannot usually be used to identify you.

The technical storage or access is required to create user profiles to send advertising, or to track the user on a website or across several websites for similar marketing purposes.

Manage options

Manage services

Manage {vendor_count} vendors

Read more about these purposes

View preferences

{title}

{title}

{title}