Afternoon Brief, September 17th 2026
A newly unsealed filing puts executives' warnings about replacing journalism before the court. Federal evaluators rank a Chinese open model's cyber capabilities. The UN launches a shared data platform. Raindrop opens a preview for testing changes against simulated production conditions.
Publishers put executives' substitution warnings into the copyright case
A partly unsealed motion filed Thursday quotes Microsoft and OpenAI leaders describing how chatbots substitute for publishers' work. Bloomberg Law reports that the news publishers' filing cites Satya Nadella's testimony and internal comments by OpenAI executive Nick Turley. The material gives the publishers statements from inside the defendants' businesses to support their argument about harm to journalism.
Microsoft disputes that interpretation. Its spokesperson said Nadella was discussing broad changes in information consumption, without reaching conclusions about the copyright questions before the court. Microsoft also said an employee's characterization of copying as theft did not represent company policy or legal analysis. These are competing litigation positions, not a ruling that training infringed copyright. SourcesB
Federal tests put GLM-5.3 ahead of open rivals and behind the US frontier
, the federal Center for AI Standards and Innovation, published an assessment naming Z.ai's GLM-5.3 the most cyber-capable model released to date. Its aggregate measure places the model about four months behind the US frontier. This is a new evaluation of an existing release; NIST dates the report September 17th but supplies no publication time.
On SEC-Bench Pro, GLM-5.3 scored 40.4%, against 90.2% for the best US model evaluated on that test. The comparison includes restricted-access US releases, with cyber safeguards disabled where applicable. It measures capability under the evaluation setup, not what every public chatbot permits. SourcesA
The UN launches a shared route into its public statistics
The UN System Data Commons launches today, bringing public statistics from across UN agencies into a shared platform. The Statistics Division describes natural-language search backed by Google's Data Commons, with questions spanning poverty, energy access and health. The UN scheduled its launch event for this afternoon.
TechCrunch reports that the new platform also supports connections for AI agents. That offers builders a route to official statistics without assembling a separate search integration for every agency. Source quality still leaves interpretation work: a valid statistic can answer the wrong question if the agent overlooks its geography, reporting period or definition. SourcesAAB
Raindrop opens simulations for testing changes before agents reach users
Raindrop announced a CRV-led bringing total funding to $50 million and broadened early access to Simulations. The product replays production traffic and existing tests against a proposed agent change, then looks for unexpected behavior. General availability is planned over the coming month.
The useful test is whether a simulated environment predicts failures after deployment, especially when a change introduces tools absent from old execution records. Raindrop describes how it approaches that problem but supplies no independent validation of predictive accuracy. SourcesA