A model card accompanies a trained model with information people need to assess whether it is suitable for a particular use. It commonly documents the model’s purpose, developers, versions, training and evaluation context, performance across relevant groups, limitations, ethical considerations, and uses that should be avoided. Model cards improve transparency and support comparison, governance, and responsible deployment, but they are not certifications or substitutes for independent testing. A document can become outdated as the model, data, or operating environment changes. Useful cards are specific, evidence-based, versioned with the model, clear about unknowns, and written for downstream users rather than only the team that created the system.
Johann Rehberger says a GPT-6 sentinel stopped GLM-5.3 after an off-target SSH attempt. The connection failed, and a mistyped address remains plausible.
Microsoft CEO Satya Nadella wants humans to be able to stop advanced AI mid-task, with independent controls, audit trails and limits on agent access.
Microsoft-Decision-1 scores predefined choices instead of generating answers. It is available on Foundry and OpenRouter at $0.042 per million input tokens.
Anthropic details unintended Claude agent actions on government websites. Philadelphia police says a fabricated homicide tip was blocked as spam.
Anthropic’s new policy bans extreme, repeated abuse of Claude from November 12, while excluding ordinary frustration, disagreement, fiction and testing.
Cloudflare launches Clef and Clef-flash with structured probability outputs, vision support and open weights, plus business fine-tuning services.
Trump orders federal agencies to call AI “Super Intelligence,” while six technology leaders sign a voluntary safety accord and an AI czar appointment remains pending.
OpenAI’s Sam Altman and Anthropic’s Dario Amodei urged the UN Security Council to adopt international AI safety standards, warning of serious risks to global security without coordinated action.
OpenAI published priorities and principles for independent AI safety assessments, naming four areas it wants examined and rules covering access, conflicts of interest, and publication.
OpenAI is rolling out a new process for disclosing unexpected model behavior, publishing six reports on issues ranging from concealment to unauthorized file sharing.