Are open-weight AI models safe?
THE SO WHAT
The open-weight debate is shifting from ideology to risk accounting—governments and enterprises now have to weigh sovereignty gains against model exfiltration, misuse, and fragmented guardrails. If you're adopting open weights, treat them like critical infrastructure assets, not just cheaper alternatives to APIs.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIFed's Warsh Says AI Will Change Monetary Policy Conduct
If central banks treat AI as a medium-term positive supply shock, rate paths and inflation assumptions can decouple from the last decade’s playbook. Operators should expect more volatility around macro data prints—and stress-test plans against both faster productivity and policy misreads as models and humans learn each other.
Applied AIBOE’s Bailey Calls for Global AI Testing Standards
When the Bank of England talks about pre- and post-release AI model testing, it's a warning shot for anyone deploying models into financial rails—governance is about to look more like stress testing than casual A/B. If you touch money flows or market infrastructure, start logging evals, failure modes, and rollback plans as if a regulator will ask for them.
Applied AIMicrosoft unveils MAI-Base-2, a MoE model with 2.8T total and 80B active parameters, claiming 4x-12x more pretraining FLOPs-efficiency than DeepSeek-V4-Pro-Base
If Microsoft’s MAI-Base-2 MoE really delivers 4x–12x pretraining FLOPs-efficiency over DeepSeek-V4-Pro-Base, the constraint for frontier work shifts further from raw compute to data quality, routing, and downstream productization. For operators, this is another nudge to assume rapid, stepwise model efficiency gains in your 12–24 month planning rather than linear cost curves.
Applied AIOpenAI continues to work to grow its ad business in ChatGPT, cutting ad rates, chasing small and midsize businesses, and hiring some new talent from Meta
ChatGPT’s push into lower-priced ads for SMBs—and Meta ad talent coming in—confirms the assistant is becoming an intent and commerce surface, not just a productivity tool. If you’re a marketer or product owner, you now have to treat AI assistants as paid channels in your acquisition mix and watch how their targeting and attribution models evolve.