David Sacks Says Small AI Models Challenge Closed-Model Business

David Sacks argued that specialized small AI models are challenging the business case for closed, general-purpose systems, while criticizing Anthropic’s portrayal of security risks tied to models on internet-connected devices.

In comments on Oct. 14, Sacks said Anthropic’s experiments relied on devices with debugging ports already open, external servers and a task that people had set up for the models. He said those conditions did not establish that small models can autonomously spread through real-world devices as an AI worm.

Sacks argued that conventional malware can already spread by exploiting default passwords, open ports and devices that lack security updates. In his view, small models have not substantially changed that existing threat pattern.

He also pointed to Coinbase’s use of its own fraud data to fine-tune a Qwen-3.5-9B model. Sacks said the model outperformed leading models, including Anthropic’s Opus 4.5, in related tests and reduced fraudulent transactions by about 30%.

Sacks said models optimized for specific business tasks can compete on cost, speed and practical performance. He argued that their adoption could reduce reliance on large, general-purpose AI systems and put pressure on companies that sell closed models, including Anthropic.