Organization · Research organization
AI Futures Project
A small research group forecasting the future of AI; publisher of the AI 2040 scenario, including an inference-only verification plan.
The AI Futures Project describes itself as "a small research group forecasting the future of AI, funded by charitable donations and grants". Its AI 2040 scenario has supplements on verification:
- Verification plan. Dean's plan would convert data centres into isolated inference units, copy their traffic with network taps, recompute random samples and wipe memory periodically, so that only verified inference outputs persist 1. See AI 2040 inference-only verification stack, Bandwidth limits and compartmentalization, Network taps and certifiers, Sampled inference recomputation and Memory wiping and proofs of secure erasure.
- Get involved. A companion page tracks the status and open problems of each component, and lists organizations working on verification 2.
- Covert AI projects. Halstead and Larsen discuss how a covert AI project might be hidden and detected, treating waste heat as the main constraint on hiding a facility 3; see Remote detection of data centres. They propose auditing the semiconductor supply chain upstream to reconstruct past chip production 3; see Chip registries and manufacturing records.
Implementations
- A proposed retrofit that isolates data-centre inference units, taps their front-end traffic and recomputes random samples to check that only declared inference runs.
Mechanisms
Mechanisms this organization has designed, built, evaluated or supplied.
- A verifier re-runs a random sample of an AI provider's logged queries on a trusted copy of the declared model and checks the outputs match.
- Capping or removing the network links between groups of accelerators, so that serving models still works but large training runs become impractically slow.
- Recording each AI chip's identity and owner from the fab onwards, and cryptographically fixing manufacturing records, so that chips can be accounted for later.
- Overwriting all of a device's memory in a way a verifier can check, so that nothing from earlier, undeclared work survives the wipe.
- Devices on a cluster's network links that copy and hash all traffic, so a verifier can later check sampled records against declared work.
- Remote detection locates large data centres and estimates their power capacity without site access, using satellite imagery, heat signatures and public records such as permits.
- Organizing all AI workloads in a facility into discrete, reproducible units, so that a verifier can recompute a random sample and check each one.
Publications
Sources this organization authored or published.
- CB. Halstead & T. Larsen (2026). Covert AI Projects. AI 2040. RecordCited by Chip registries and manufacturing records; Remote detection of data centres; AI Futures Project; Epoch AI
- CAI Futures Project (2026). Get Involved in Verification. AI 2040. RecordCited by AI 2040 inference-only verification stack; AI Futures Project; Amodo Design; Future of Life Institute; RAND; Singapore AI Safety Hub (SASH)
- CR. Dean (2026). Verification Plan. AI 2040. RecordCited by Bandwidth limits and compartmentalization; Deterministic and bit-exact inference; Memory wiping and proofs of secure erasure; Network taps and certifiers; Reproducible computation packets; Sampled inference recomputation; Tamper evidence for verifier devices; Timed challenge-response and memory-occupation challenges; AI 2040 inference-only verification stack; Communication between compute groups is bounded; Declared hardware is idle or shut down; This compute runs inference, not training; Sampling and assurance; Tamper evidence and tamper resistance; AI Futures Project
Sources
- CR. Dean (2026). Verification Plan. AI 2040. Source recordSupports: AI 2040 verification plan: isolated inference units, taps, recomputation, memory wipes
- CAI Futures Project (2026). Get Involved in Verification. AI 2040. Source recordSupports: companion page: component status, open problems, organizations working on verification
- CB. Halstead & T. Larsen (2026). Covert AI Projects. AI 2040. Source recordSupports: covert AI projects supplement: waste heat, detection, upstream supply-chain audits