Agentic AI Meets Cybersecurity + Solar Robots in the Desert | E2125

Key Concepts

Agentic AI security, AI agents, LLMs, Model Context Protocol (MCP), Agent-to-Agent (A2A), pentesting, offensive security, continuous testing, solar power, utility-scale solar, automation, on-site prefab, Terapab, solar power plant construction, solar panel efficiency, energy storage, grid connectivity.

Agentic AI Security with Zoe

Defining Agentic AI

• AI Agents: Defined as LLMs operating within a framework, capable of tasks ranging from sending emails to controlling physical security systems. LLM agents are considered "gamechanging" due to their far-reaching applications.
• Security Concerns: The ability of AI agents to execute tasks independently raises security issues, as these agents operate outside traditional security perimeters.

Security Risks and Vulnerabilities

• Probabilistic vs. Deterministic Models: AI models are probabilistic, making traditional testing methods less effective.
• Model Context Protocol (MCP): MCP allows models to interact with various tools and databases, expanding the attack surface.
• SQL Injection: Models interacting with databases are vulnerable to old-school vulnerabilities like SQL injection.
• Agent-to-Agent (A2A): Google's A2A standard enables agents to communicate, further complicating security.
• Internal vs. External Threats: While AI agents are often internally sourced, external interactions and MCP usage can create vulnerabilities.
• Expanded Attack Surface: Integrating AI agents expands the attack surface, creating more potential points of exploitation.

Addressing Security Gaps

• Chasing the Problem: Cyber defenses often lag behind the development of AI solutions.
• Governance and Security Teams: Security teams are often playing catch-up, trying to secure AI solutions after they've been developed.
• Owasp and MITRE: Organizations like OWASP (Open Worldwide Application Security Project) and MITRE provide information on potential vulnerabilities. MITRE Atlas is a framework for identifying vulnerabilities in AI deployments.

Pentesting and Offensive Security

• Validation Needed: Validation is crucial, and offensive security is the best approach.
• Continuous Pentesting: Due to the rapid advancement of LLMs, continuous pentesting is necessary (weekly or daily).
• Holistic Testing: Pentesting should encompass the entire environment, including social engineering aspects.

Market Adoption and Future Outlook

• Executive Interest vs. Security Concerns: Executive teams are pushing for AI adoption, but security concerns are slowing down deployment.
• Limited Agent Onboarding: Most organizations have only onboarded a few AI agents due to security concerns.
• Future Agent Deployment: The expectation is that enterprises will deploy thousands of agents in the coming years.
• Timeline for Agentic Era: Optimistic estimates suggest the floodgates will open in 6 months, but widespread adoption depends on solving security issues.
• Multi-Agent Systems: Multi-agent systems, where agents interact, further complicate security testing.
• AI-Powered Pentesting: Zoe's solution uses AI to test other AI, interrogating them to extract information.
• AI Police: The concept of AI policing other AI is emerging, with AI being trained to identify and mitigate bad actor tendencies.
• Treating Agents Like Humans: AI agents should be treated like human employees, with evaluations of their work.
• Future Focus: Expanding security purview to the engineering side, ensuring code produced by AI is checked by AI.

Zoe's Business and Hiring

• Target Customers: Larger companies in highly regulated industries like finance.
• Partnerships: Partnering with hyperscalers to offer innate security services.
• Hiring Needs: Security analysts with pentesting experience and full-stack engineers with Rust and Typescript expertise.

Terabase and the Future of Solar Power

Overview of Terabase

• Focus: Building digital and automation tools to design, build, and operate utility-scale solar projects more efficiently.
• Utility-Scale Solar: Giant solar power plants in remote areas, spanning thousands of acres (100 MW to multi-GW).
• Three Key Areas: Design, construction (Terapab), and operation.
• Synthetic Vertical Integration: Software platform that integrates the functions of developer, contractor, and operator.

Solar Power Progress in the US

• Banner Year: 2024 was a banner year for solar installations in the US.
• New Generation: Solar accounted for 66% of new generation capacity in the US.

Terapab: On-Site Prefabrication

• Problem Statement: Automating the construction of solar plants.
• On-Site Prefab Facility: A pop-up factory (200-foot assembly line) set up on-site to preassemble components.
• Production Rate: Can produce 1 MW in 8 hours.
• Mobility: The Terapab facility is moved around the site as sections of the solar farm are completed.
• Human-Robot Collaboration: Humans and robots work together, with humans handling fine motor tasks.
• Labor Productivity: Demonstrated labor productivity improvements of 25%.

Remoteness and Electricity Transference

• Remote Locations: Solar projects are typically located in remote areas, often hours away from major cities.
• Electricity Loss: Electricity loss during transference is around 6-9%, depending on distance and voltage.

Goals and Objectives

• Speed: Accelerating the pace of solar deployment.
• Cost: Lowering the cost of solar construction.
• Quality: Ensuring high-quality construction for long-term operation (40-50 years).

Solar Panel Lifespan and Efficiency

• Long Lifespan: Solar power plants are designed to operate for decades (30-40 years).
• Panel Replacement: Panels may be replaced with more efficient models in the future.
• Efficiency Limits: Current silicon-based solar cells are approaching their efficiency limits.
• Next-Gen Semiconductors: Perovskite semiconductors could increase efficiency from 22% to 35-40% in the next 10 years.

Energy Storage

• Storage Integration: Working closely with storage companies and developing software to manage the interaction between the solar farm and the grid.
• Plug-and-Play Storage: Storage systems are often containerized and plug-and-play.

Off-Grid Solar and Data Centers

• Off-Grid Potential: The future may involve off-grid solar systems, where data centers are powered by standalone solar farms with battery backup.
• Island Basis: Multi-gigawatt systems operating on a standalone basis.

Costs and Economics

• Solar Cost: Approximately $1 per watt for solar installations in the US.
• UAE Project: A 5.2 GW solar project with a 19 GWh battery and 1 GW of 24/7 solar output costs around $6 billion.
• Nuclear Comparison: A 1 GW nuclear power plant would cost $15-20 billion and take 15-20 years to build.

Terapab V2 and Future Development

• Terapab V2: A fully automated system that will double the speed of the current generation.
• Cost Reduction: Aiming to cut the cost of solar deployment in half.
• Inspiration: Drawing inspiration from SpaceX's approach to reusable rockets and incremental improvements.

Growth and Future Plans

• Growth Target: Aiming for 100% year-over-year growth.
• IPO: Considering an IPO in the future (potentially 2027).
• Global Expansion: Active in the US, Australia, Europe, and the Middle East.

Hiring Needs

• Programmers and Roboticists: Actively seeking talented programmers and roboticists.

Synthesis/Conclusion

The podcast episode features two interviews highlighting innovative approaches to pressing global challenges. Zoe addresses the emerging security risks associated with the increasing adoption of AI agents, emphasizing the need for continuous testing and proactive security measures. Terabase focuses on accelerating the deployment of solar power through automation and on-site prefabrication, aiming to reduce costs and increase efficiency. Both companies are leveraging technology to address critical needs in their respective fields, with significant potential for growth and impact.