Research Note: Sealed Computation, in Practice
- Published
- Authors
- Paul Bricman
Proof of concept for demonstrating how sealed computation can be used to prove that an inference workload was executed in a monitored setting.
Read moreResources, announcements, and opportunities on addressing emerging challenges posed by artificial intelligence.
Proof of concept for demonstrating how sealed computation can be used to prove that an inference workload was executed in a monitored setting.
Read moreWe are excited to share that we have line of sight on technology for enabling the verifiable monitoring of inference workloads.
Read moreHow can we prove that an AI output was produced at a particular time and place, using a particular GPU, all without having knowledge of the actual workload?
Read moreWe reverse engineer the memory layout of an accelerated inference process by intercepting kernels and injecting custom assembly for activity monitoring.
Read moreTechnical maturity is necessary, but not sufficient, for enabling virtual diplomacy. We lay out a path for ensuring its political maturity as a key factor in its diffusion.
Read moreStrengthen state capacity to wield AI as a geopolitical resource.
Read moreThe mechanics of power in the age of "countries of geniuses in a datacenter."
Read moreWe carried out predeployment testing of OpenAI's o3-mini model using our Cybersecurity Simulator, a procedural cyber range optimized for realism, variety, and granularity.
Read moreAutoHack is a platform that offers offensive security challenges modeled after real-world cybersecurity kill chains. We're running an inaugural private tournament with a $10K prize pool.
Read more