Which AI labs commit to third party vetting by October 31?

The key fact is that Google has not yet issued a public commitment that clearly matches Anthropic’s standard of ongoing, employee-level third-party evaluator access to systems, models, and training processes. The market is already reflecting that gap: traders appear to assign meaningful but limited odds, which is consistent with a story where the idea is live in the industry discussion but has not crossed the threshold into an official company pledge. That matters because this market is not about general support for evaluation, pilot access, or pre-release testing; it requires an explicit commitment that is broader, deeper, and more durable than a normal launch review or a customer-facing evaluation program.

The recent context is mildly supportive of a Yes outcome only in the sense that the conversation around external evaluation has become more prominent. Anthropic’s move raises the reputational bar for other frontier labs, and there is evidence that some industry voices are increasingly comfortable with stronger evaluator access. Still, the available signals around Google point mostly to model development, launch timing, infrastructure work, and general evaluation-related services rather than a concrete promise to open up internal systems to outside evaluators on an ongoing basis. A company like Google could decide to remain more selective, especially if it believes its existing safety, red-teaming, and internal review processes are sufficient or if it wants to avoid exposing training workflows and sensitive model development details.

From a market perspective, the current price implies the crowd sees some chance of a late policy shift, but not a high one. That seems reasonable because the remaining window is short and the required announcement must be official and unambiguous. Companies often prefer to use carefully scoped language around evaluation, access, and safety commitments, and the difference between a broad endorsement of third-party evaluation and a qualifying commitment is material here. Unless Google makes a targeted announcement that mirrors Anthropic’s language closely, the default expectation should remain No.

The main bullish case is that Google may feel pressure to avoid looking less transparent than competitors, especially if external evaluators and policymakers continue pushing for more access. But the strongest evidence today is still the absence of a qualifying statement, which is especially important given the market’s strict resolution rules. For Yes to happen, Google needs to do more than discuss evaluation in principle; it needs to publicly commit to ongoing third-party access in a way that clearly covers the development process.