talks

Mythos, LLMs, and Implications for Security in the Era of AI and Autonomy

How LLM capabilities change the threat model, and what that implies for defenders.

Putting the AI in AIOps: From Artificial Intelligence to Actionable Insights

A tour of the AIOps stack — data, modeling, and workflows — and where the real bottlenecks are.

also given at Cloud Control Workshop (June 2025)

Creating Scientific Foundations for Cyber Autonomy

Security operations still run at human timescales. As AI-driven offense improves, that becomes untenable — this talk argues for an interdisciplinary science of cyber autonomy spanning AI, security, systems, formal methods, and policy.

also given at Snowflake

Advice on Demos and Presentations

A cartoon-driven take on what makes a demo or a talk land.

Incalmo: An Autonomous LLM-Assisted System for Red Teaming Multi-Host Networks

Incalmo plans red-team exercises as high-level declarative tasks executed by domain-specific agents; it acquires critical assets in 37 of 40 environments.

Science and Sip

Word puzzles and trivia, no research content whatsoever.

Toward Cyber Autonomy for Securing High Performance Computing Environments

HPC infrastructure is a prime target for exfiltration, credential theft, and compute hijacking. Presents Incalmo and MHBench as first steps toward autonomous defense.

"Meta" Guidelines for Research

The intangible part of the research process: choosing problems, formulating them, and communicating results.

also given at AINTEC 2025; CMU ECE PhD orientation (2024)

Synthetic Data Generation for Accelerating Networked Systems Innovation

Joint work with Giulia Fanti, Zinan Lin, Yucheng Yin, Steven Dong, and Lesley Zhou.

also given at LDOS, Austin (June 2025)

Future of Enterprise Security Initiative: Concluding Update

Wrap-up of the CyLab initiative co-directed with Lujo Bauer.

Cyber Autonomy @ CyLab

The case for a CyLab initiative on cyber autonomy, co-directed with Lujo Bauer.

also given at CIO Summit (December 2025)

Enabling Software-Defined Network Security for Next-Generation Networks

Incalmo: An Autonomous LLM-Assisted System for Red Teaming Multi-Host Networks

Extended version of the Incalmo talk with more background on the attack pipeline.

Some Lessons from a "Long" Career in Academia

Reducing Cloud Costs for Metrics Observability by 50x

Sketch-based precomputation for high-cardinality metrics, with a Prometheus-compatible implementation. With Milind Srivastava and Alan Liu.

Enabling AIOps for Next Generation Networks

A 20,000-foot view of AIOps: data, analytics/modeling, and AI workflows.

Reflections on the Research/Product Design Process

Enabling Data-Driven Innovation with Synthetic Data

Enabling Data-Driven Innovation in Product Lifecycles with Synthetic Data

Joint work with Giulia Fanti, Zinan Lin, and Yucheng Yin (Project DataFuel).

Unlocking AI/ML Innovation in Healthcare Using Synthetic Data

Mastering the Innovation Process in Product & Research

Guiding Principles for Research/Product Engineering

Opportunities and Challenges of Synthetic Data for Enterprise Workflows

Credits: Giulia Fanti, Zinan Lin, Yucheng Yin.

A Platform for Mitigating Data Bottlenecks with Synthetic Trace Data Generation

Synthetic Data to Enable Collaboration and Data Science for Networking

Joint work with Giulia Fanti, Zinan Lin, and Yucheng Yin (Project DataFuel).

Enabling Data-Driven Collaborative System Workflows with GAN-Based Synthetic Data Generation