How Two Sigma Gives AI Agents Safe Identities

AI Engineer21m 9s
0 comments · 0 votesOpen discussionClose discussion
Sign in to join the discussion

    Video summary

    Shu Fang explains the security problem created when cloud agents act through a human user's long-lived credentials. Shared identities blur who performed an action, make permissions too broad and weaken revocation and audit controls.

    Two Sigma's approach treats each agent session as a separate principal. A broker issues short-lived credentials, policy limits what the agent can reach and logs connect the resulting actions to both the agent and the person who launched it.

    Fang calls these agents tethered because they remain linked to an accountable user without impersonating that user. The design supports least privilege and clearer incident response while preserving the productivity benefits of remote autonomous work.

    Original YouTube thumbnailWatch on YouTube

    Share this page

    Shu Fang beside the headline Agents Get Safe Identities Framed in blue with WWW.ARTIFICIAL-INTELLIGENCE.VIDEO, 3 September 2026 and duration 21m 9s.

    Shu Fang argues that remote AI agents need their own short-lived identities and scoped permissions so organizations can authorize, audit and revoke agent actions without sharing a user's credentials.