What strategies are most effective for securing AI workloads?
The answers emphasize restricting AI agent permissions, securing environments and data sources, and implementing strong runtime and network controls to prevent unauthorized access and unpredictable behavior.
6 answers
Answers


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.
Latent Space · Watch on YouTube

Lou E.
Federal CTO · ColorTokens Inc.
Limit Asset Permissions
Lou E. says organizations over-permission AI agents and argues they should focus more at the asset level so agents cannot jump out of their sandbox, such as an HR agent reaching the finance server.
Fed Gov Today · Watch on YouTube

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.
The Stack Overflow Podcast · Listen on Apple Podcasts

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.
The Control Layer with Amer Altaf · Listen on Apple Podcasts

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.
It's About Data · It's About Data · Listen at 13:11
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.
CISO Talk by James Azar · CISO Talk by James Azar · Read
More Information Technology (IT) questions
View all →- What practice secures AI agents in enterprise environments?6 answers
- What is the most important governance for enterprise AI products?6 answers
- What safeguards matter most for protecting data with AI tools?5 answers
- What is your first step in modernizing IT with AI?6 answers
- What is the top security measure for enterprise AI deployment?6 answers














