Bloomberg AI engineer Andrey Rybka warned that traditional sandboxes fail against autonomous AI attacks. Organizations must isolate agent identities, block outbound network lookups, and require human approval for high-risk actions.
FireBrief
FireBrief