We scan new podcasts and send you the top 5 insights daily.
Tom Milton says a government could require a zero-knowledge proof that a safety classifier is running on each API response. Then it could send random requests and check the proof remains attached, without inspecting the lab's hardware. This gives a correctness claim, though not proof no illegal model runs elsewhere.
Instead of trying to anticipate every potential harm, AI regulation should mandate open, internationally consistent audit trails, similar to financial transaction logs. This shifts the focus from pre-approval to post-hoc accountability, allowing regulators and the public to address harms as they emerge.
Tom Milton argues verification is not only for outside regulators. If AI models gain nation-state-level cyber capability, labs need hardware-backed logs to confirm models have not changed their own weights or spoofed evaluation results. The same stack protects against untrusted AI as against an adversary country.
With AI incidents rising and safety benchmarks lagging, the era of "trust me" AI governance is ending. The podcast hosts predict that the market will soon demand exportable proof and certifications (like SOC 2 for AI) from vendors before deploying their systems, shifting the impetus for safety from regulators to customers.
Tom Milton says the fallback is turning data centers off, but that is extremely costly. Verification lets a prover show it is serving only inference, or running small safety experiments, so a slowdown can preserve useful work while stopping dangerous training. This reduces the cost of reaching a deal.
For AI safety, Demis Hassabis advocates for an international regulatory body, similar to the International Atomic Energy Agency. This body would have technical experts who audit frontier models against agreed-upon benchmarks, checking for undesirable properties like deception and ensuring public confidence through independent verification.
Instead of supervising an AI's hidden thought process, we can demand it produces a 'certificate of reasoning'—a checkable proof—along with its output. This could include citations or sensitivity analyses, shifting verification from observing the process to checking the provided proof.
Confidentiality concerns make AI companies hesitant to admit external human auditors into proprietary codebases and training runs. Highly capable AI auditors whose internal memory states are securely wiped immediately after compiling redaction-adjusted compliance reports can verify safety without leaking intellectual property.
Frontier AI labs have deep technical knowledge but also an incentive to ship products, while governments have national security concerns but lack expertise. This creates a trust gap, necessitating a neutral third party—like a Moody's for AI—to perform technical audits and provide trustworthy risk assessments.
The goal for trustworthy AI isn't simply open-source code, but verifiability. This means having mathematical proof, like attestations from secure enclaves, that the code running on a server exactly matches the public, auditable code, ensuring no hidden manipulation.
Tom Milton says recomputation requires the verifier to access model weights inside the prover's data center, which labs dislike. Zero-knowledge proofs would let provers demonstrate correct inference without sharing weights, though current ZKML is orders of magnitude slower; sampling one in a thousand packets could suffice.