🐉 BSidesSF 2025: Cloud Security in The Age of AI

Sreeram Sandrapati recently attended BSidesSF, where he explored how agentic AI and evolving cloud architectures redefine security strategies across the industry. Key concerns covered at the event included securing large language models (LLMs) handling sensitive data, addressing prompt injection and model poisoning, and adapting frameworks for AI-specific threats. The following is a summary of a longer article Sreeram wrote, which is posted on Medium.

Key Insights from BSidesSF

Centralizing Security Across Hybrid Environments

Implementing consistent security controls across environments with different native security models offers several benefits:

• Consistent policy enforcement and unified traffic management
• Comprehensive monitoring reducing data leak risks
• Dynamic threat mitigation capabilities

TrustLogix's platform enables no-code policies that can be written once and deployed everywhere, allowing for near-instantaneous updates and real-time risk monitoring.

New Attack Surfaces in Cloud Environments

Cloud environments have expanded attack surfaces beyond traditional perimeters:

• Cloud service provider endpoints and Kubernetes API servers are becoming targets
• IAM privilege escalation paths unique to cloud environments

Traditional indicators of compromise are insufficient, necessitating behavioral indicators. TrustLogix mitigates these risks through the automated inventory of data stores, behavioral baselining to flag deviations, and fine-grained enforcement at the data layer.

Securing AI Agents

AI agents face unique vulnerabilities, particularly Authorization and Control Hijacking through direct exploitation, permission escalation, and role inheritance manipulation. Key attack scenarios include:

• Permission Window Abuse: Exploiting temporary access by manipulating task queues
• Role Chaining Exploitation: Sequencing legitimate requests to create unauthorized access

TrustLogix addresses these with agent-specific credentials, just-in-time entitlements, and immutable audit trails.

Future Trends

• "Secure by Default": AI development speed necessitates built-in security controls
• "Vibe Coding": Tools needed for securing AI-generated code with proper verification
• Shifting Security Focus: Attack vectors moving from applications to CI/CD pipelines

Conclusion

BSidesSF 2025 revealed an industry at an inflection point where AI, cloud architectures, and human factors create challenges and opportunities. Key principles include specialized security for AI systems, centralized visibility across environments, and the continued importance of human psychology in security.

For a more in-depth look at the topics and how TrustLogix can help with these issues, please read the full blog: 🐉 BSidesSF 2025: Focusing on the Evolution of Cloud Security in The Age of AI