RESEARCH · GLOBAL
Anthropic study raises cyberattack risk from open AI models; industry debates tradeoff
Anthropic published research arguing that open-weight LLMs (like GLM-5.3) pose material cybersecurity risk due to ease of misuse, sparking debate on benefits (research access, model diversity) vs. safety (attack surface for financial crime, fraud).
WHY IT MATTERS
For BFSI, direct tension: open models enable faster AI adoption and reduce vendor lock-in, but increase attack surface. Regulators (FCA, OCC) will likely tighten controls on open-model use in production for sensitive workflows.