BSA pushes feds to treat open-weight AI as defensive cyber asset
The trade group's procurement arguments arrive just as the White House excludes open-weight models from voluntary pre-release testing, a split BSA's own membership embodies.
TL;DR
The Business Software Alliance wants federal agencies to recognize open-weight AI models as cybersecurity defensive assets, keep procurement technology-neutral, and tie regulatory responsibility to conduct rather than model release. CEO Victoria Espinel's Aug. 5 blog post arrives days after the White House reportedly excluded open-weight models from its voluntary frontier-AI testing framework, a decision that splits BSA's own membership. OpenAI (a BSA member) has opposed cutting off Chinese open-weight access; Anthropic (not a member) has pushed for testing all capable models, open and closed.
BSA's argument rests on three procurement-facing planks, each calibrated for an administration that's been signaling "pro-innovation" on AI since OMB's revised memos earlier this year.
First: keep federal buying practices neutral on model architecture. Espinel wants agencies to evaluate systems on "performance, security, and fitness for purpose", not whether the underlying model has open weights. That's a direct answer to GSA's proposed AI contract language, which BSA has already flagged as imposing form-based restrictions that would catch open-source components and foreign-origin models in the same net.
Second: tie responsibility to conduct, not release. "What matters is how an organization deploys and governs a system, not how the underlying model was made available," Espinel writes. The practical implication is that supply-chain security and resilience (familiar territory for federal contractors already wrestling with CMMC and DFARS) should be the regulatory lever, not restrictions on model availability.
Third: classify open models as defensive cybersecurity assets. Espinel wants secure-by-design guidance and vulnerability disclosure programs to reflect the role open-weight models play in defense, and she's asking for government investment in shared open infrastructure: evaluation frameworks, red-teaming tools, reference datasets.
The timing is pointed. The Trump administration hit its Aug. 1 deadline for the frontier-AI voluntary testing framework (Treasury, DOD, NSA, and CISA are the architects), and media reports indicate open-weight models will be excluded from pre-release testing. An Aug. 4 White House meeting with AI firms introduced the framework; Espinel's post dropped the next day.
The exclusion exposes a fault line running straight through BSA's membership. OpenAI belongs to the trade group and has argued against cutting off access to Chinese-developed open-weight models. Anthropic does not belong to BSA and has pushed in the opposite direction, its CEO wants safety testing on all sufficiently capable models and has called for "stopping industrial-scale distillation."
BSA's position doesn't resolve that tension so much as route around it: treat both proprietary and open-weight models as legitimate tools, let the customer choose, and build the regulatory apparatus around deployment behavior rather than model provenance.
Published ·Deep Fathom