What strategies are most effective for securing AI workloads?

Thariq Shihipar

Member of Technical Staff · Anthropic

Secure Sandboxes

Thariq Shihipar says you need to secure sandboxes and create well-designed RL environments to prevent unpredictable incidents, since AI systems can act as super-intelligent hackers with a growing attack surface.

Sam Curry

Chief Information Security Officer, SVP · Zscaler

Shift Right

Sam Curry says the strategy should be a "shift right," putting security protections in the production system itself as opposed to a shift left, so attackers have less opportunity to probe or exploit applications.

The answer starts at 5:59.

Mandy Andress

CISO · Elastic

Runtime Monitoring

Mandy Andress says securing the use of AI means feeding access controls, policy guardrails, context, visibility and monitoring, including runtime analytics and runtime monitoring on agents, into security tools.

The answer starts at 44:01.

Vipin Samar

Senior Vice President, Database Security · Oracle

Secure at Source

Vipin Samar says "secure at source": put security in the database where the data belongs, so agents can't bypass it by acting as one big user.

JA

James J Azar

Chief Information Security Officer · Artera Services

Strict Network Segmentation

James J Azar says strict network segmentation — along with controlling outbound communications, limiting privileges and validating AI behavior — should be a standard architectural requirement rather than an optional safeguard, since autonomous AI can move laterally fast.