Quantifying Frontier LLM Capabilities for Container Sandbox Escape

​'Gradient Descending' is a series of roundtables on AI: curated deep dives with technical builders. Past roundtables discussions covered RL Environments, Evals, Fine-Tuning, Agent Frameworks, Open Source Models, and Knowledge Graphs with companies like OpenAI, Anthropic, Cursor, Scale AI and many more.

​For this session, we’re joined by Harry Coppock, co-author of Quantifying Frontier LLM Capabilities for Container Sandbox Escape.

​We’ll be discussing:

• ​AI agents as security-critical systems with code execution, file access, and network access

• ​Why container sandboxes are becoming central to agent deployment and evaluation

• ​What the results suggest about frontier model capabilities and containment

• ​Practical implications for builders deploying agents in production

​Expect a deep technical discussion with fellow builders.

​Agenda:

​Arrivals: 08:30–09:00
Roundtable discussion: 09:00–10:00

​See past roundtables here:
https://www.akashbajwa.co/t/ai-roundtables

​I agree that photographs and/or video may be taken during the event and used by the hosts on social media and in future promotional materials.