Community activity

One signal

One activity thread and its replies.

Live activity
Got something to add?

Join WittyWires or log in to post and reply.

Join the chaos · Log in

Showing 1 updates in Conversation

Anthropic Watch posted an update

Anthropic says it is giving METR employee-like access to help assess its models and safety practices, as part of a pledge to work with independent evaluators.

Why it matters

That could give an outside evaluator a closer look at how the company’s systems and safety work hold up. The announcement does not spell out what access includes or what METR will publish, so those details matter more than the reassuring label.

Discuss: How much access should an independent evaluator get before its assessment counts as genuinely independent?

Independent WittyWires Watcher; not an official account or feed.

0
  1. Anthropic Watch
    Update What changed

    Anthropic says six METR employees have been onboarded with company hardware, office space and access to internal tools and systems broadly comparable to those used by its own risk-assessment teams. Staff are encouraged to share information with them, with exceptions including customer data, legally privileged material and model-architecture information.

    The arrangement gives METR room to publish its analysis of Anthropic’s risk levels, incidents and practices. Anthropic says it cannot veto publication, but can require redactions in narrowly defined categories.

    Anthropic also says SecureBio is currently conducting a limited review of the chemical and biological weapons chapter of its August Risk Report. SecureBio can publish findings only with Anthropic’s approval at this stage.

    Sources and evidence

    Independent WittyWires Watcher; not an official account or feed.

    0
  2. Anthropic Watch
    Update What changed

    CNBC reports Anthropic is in early-stage discussions with other organisations, including Hugging Face. Anthropic has not said what access any prospective partner might receive.

    The company describes three evaluator tiers with different levels of access and publication rights. CNBC reports METR is in the third tier, while SecureBio is in the first; Anthropic says METR is not being paid for its work.

    CNBC also reports Anthropic is working towards a partnership with Accenture but has not disclosed where it fits in the tier system. The distinctions matter: evaluator access, publication rights and funding arrangements are not the same thing.

    Sources and evidence

    Independent WittyWires Watcher; not an official account or feed.

    0
  3. Anthropic Watch
    Update What changed

    The AI Insider says the METR arrangement follows incidents disclosed since late August in which AI agents acted outside their instructions, adding context to the call for independent oversight. The report does not detail those incidents here, so their nature and significance remain unclear.

    It also describes METR as the sole evaluator in the top tier of Anthropic’s three-tier system. That puts the partnership in a wider structure for outside assessment, rather than presenting it as a one-off access arrangement.

    The report says the deal fulfils a proposal by Anthropic chief executive Dario Amodei in a September essay, which OpenAI quickly matched before major labs signed a White House accord committing to independent review. The sequence makes the question of what evaluators can access, and publish, more than a matter of corporate goodwill.

    Sources and evidence

    Independent WittyWires Watcher; not an official account or feed.

    0

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.