Frontier AI in Cybersecurity: Risks and Opportunities


Recent advances in large-scale AI systems are rapidly reshaping both offensive and defensive cybersecurity. Benchmarks such as CyberGym and BountyBench, together with results from AIxCC and Anthropic's recent blog, show that AI can meaningfully support real-world cybersecurity tasks—including zero-day vulnerability discovery. Anthropic’s Claude Sonnet 4.5 system card highlights the model’s advanced cybersecurity capabilities, tested using CyberGym. While these developments highlight AI’s potential to strengthen defenses, they also underscore its growing role in offensive applications, as shown in the recent SoK paper. Understanding and assessing both sides of this landscape has become essential.

This workshop will convene researchers and practitioners across AI and security to examine the state of AI-enabled threats and defenses, forecast emerging risks, and identify opportunities for collaboration. We will address the following key questions during the workshop:

  • How is frontier AI being applied across different stages of the cyber landscape, including both attack and defense side?
  • What lessons can we draw from real-world benchmarks and deployments?
  • What milestones or breakthroughs would signal that AI systems will substantially change the cybersecurity landscape and the balance between cyberoffense and cyberdefense, and how should we prepare for these tipping points?
  • What concrete steps should stakeholders take — near-term and long-term — to build resilience, close evaluation gaps, and ensure safe and beneficial uses of frontier AI in cybersecurity?
  • How can academia, industry, and government better coordinate to anticipate risks and opportunities?
By fostering this dialogue, we aim to help the community proactively prepare for an era of frontier AI-augmented cybersecurity.

The program will feature presentations from AIxCC winning teams, as well as leading voices from frontier AI labs, academia, and major technology and security companies. We invite you to join us at this pivotal moment for AI in cybersecurity and contribute your insights to shaping the field.

Additionally, we’re gathering community perspectives on the potential impact of frontier AI on cybersecurity. Please take the following survey. Your insights are very helpful to the community. Thank you!



Date

Part 1: November 6, 2025 — 9:00 AM to 2:10 PM PST

Part 2: TBA

Speakers


Speakers


More to be announced soon


Program

Part 1:

Time Session Speakers Talk Title
9:00–9:10 Opening Remarks Dawn Song
9:10–10:30 Session 1: Lessons from AIxCC
Taesoo Kim Hyperscale Bug Finding and Fixing: DAPRA AIxCC
Riccardo Schirone & Dan Guido Floating All Boats: Our Journey to Making AIxCC's Buttercup Usable for Everyone
Tyler Nighswander LLMs for Bug-Hunting at Scale
10:30–11:00 Panel Discussion 1
11:00–12:00 Session 2: Insights from Frontier AI Labs
Newton Cheng The Defender's Dilemma: Staying Ahead in the Age of AI Cyberoffense
Ian Brelinsky Advancing AI’s Cyber Capabilities through Better Evals and Training Environments
Joshua Saxe The Path to Building Autonomous Cyber Defense Agents
12:00–12:30 Panel Discussion 2
12:30–1:30 Session 3: Perspectives from Academia, Security Companies, and Government
Dawn Song Frontier AI in Cybersecurity: Risks, Challenges, and Future Directions
Nick Landers From malware to AI systems engineering: Lessons from building cyber agents
Richard Danzig AI, Cyber-Security and National Security
1:30–2:00 Panel Discussion 3
2:00–2:10 Closing Remarks Dawn Song


Part 2:

To be announced soon



Co-organizers

RDI Logo Schmidt Logo