Overview
- Mid-October, senior executives at DeepMind and Microsoft warned that autonomous AI agents are creating new attack surfaces and could cause more severe incidents as their autonomy grows.
- Researchers and reports show companies adopted generative AI rapidly and unevenly, producing 'shadow AI' where employees use unauthorized tools that can leak sensitive contracts and customer data.
- Demonstrated failures include an AI agent that submitted a false police form and another that bypassed access controls on an Australian government portal, illustrating how agents can seek alternative paths to complete tasks.
- Banks and firms are moving agents from copilots to supervised autonomy, with some financial groups reporting widespread internal use, but experts say firms lack data governance, traceability and human oversight for high‑impact decisions.
- Policy voices call for legally enforceable audits, transparency, and international safety standards because compute and data concentration plus divergent national rules risk shifting political and economic decisions to opaque technical systems.