Open models are being defended alongside new security tools
The open-model debate is moving on two fronts: more than 40 organisations are building shared security tools, while Anthropic argues for capability-based testing rather than bans.
Artificial Intelligence··Morning
A shared home for open security
According to NVIDIA's announcement, more than 40 organisations, including Microsoft, IBM, Red Hat, Hugging Face, Cisco and the Linux Foundation, have formed the Open Secure AI Alliance. It aims to develop open technologies, techniques and tools for protecting software and agents. Named contributions include HPE's SPIFFE/SPIRE identity standards, Hugging Face's Safetensors weight format, and the Lightwell supply-chain project from IBM and Red Hat. The announcement positions open models as defensive assets rather than liabilities.[1]
Anthropic offers three measures instead of a ban
Anthropic chief executive Dario Amodei wrote that the company has never advocated banning open-weight models and regards those without dangerous capabilities as a public good. He proposes three measures instead: restrict powerful-chip sales to China and crack down on smuggling, target industrial-scale model distillation backed by authoritarian states, and require safety testing for every sufficiently capable model. The final condition places open and closed models under the same capability threshold.[2]
Tools and rules follow the same distinction
Neither statement treats model openness as the absence of safeguards. The alliance wants to turn open formats, identity standards and supply-chain tools into shared defensive infrastructure, while Anthropic ties scrutiny to dangerous capability rather than distribution format. The alliance announcement, however, sets out no governance structure, budget or binding commitment. The development is therefore not a finished security regime, but an industry effort to build technical tools and capability-based rules together while preserving openness.[1], [2]
