31 October 2026 · Who decides how AI is used?
Publishers wanting a say over their content, researchers observing chatbots’ political answers, businesses testing assistants without customer data. Today’s selection follows these developments and ends at a radio telescope, where agents are also beginning to receive operational tasks.
OpenAI · A shared signal for publishers’ choices
OpenAI says it will honour a shared robots.txt signal for preferences on content use in generative AI training. The aim is to reduce separate rules for each participating provider. Implementation is planned over the coming months: this is not an already active universal protection. Existing crawler controls remain in place.
OpenAI · training preferences, 30 October 2026 ↗MIT · Voting answers come under observation
MIT presents the LLM Election Observatory, comparing chatbot answers about the 2026 US midterm elections. Questions are repeated with different contexts and simulated user profiles. The longitudinal analysis is unfinished: differences in tone and cited sources alone do not demonstrate an effect on actual voting.
MIT News · chatbots and elections, 30 October 2026 ↗Microsoft · Simulated businesses put Copilot to the test
Microsoft Research describes synthetic enterprise environments with artificial emails, meetings and documents to evaluate Copilot without real customer data. Tests reproduce pharmaceutical, capital-markets and software workflows. These are company-designed tests: they support repeatable evaluations but do not guarantee identical results in every real organisation.
Microsoft Research · evaluation with synthetic data, 30 October 2026 ↗NVIDIA · Agents enter the radio observatory
NVIDIA describes the use of Holoscan at the Allen Telescope Array with Breakthrough Listen and the SETI Institute. Agents can assist with instrument operation and signal analysis. The blog reports real-time AI detection of a pulsar in the Crab Nebula: a technical result, not the discovery of extraterrestrial life.
NVIDIA · agents and radio telescopes, 30 October 2026 ↗The desk’s view · Conditions matter as much as capabilities
Our connecting theme is verifiability: who authorises content use, how an answer changes and which data supports a test. Before turning an announcement into a certainty, open the source and distinguish what is operational from what remains to be demonstrated.
Sources and documents
Document date or reference period: Original sources published on 30 October 2026, selected for the 31 October edition. Not all announcements fall within the Italian overnight period..
- OpenAI · training preferences, 30 October 2026 ↗
- MIT News · chatbots and elections, 30 October 2026 ↗
- Microsoft Research · evaluation with synthetic data, 30 October 2026 ↗
- NVIDIA · agents and radio telescopes, 30 October 2026 ↗
Desk selection and summary with AI assistance, based on the four linked original sources. Manufacturer announcements are attributed to the companies; ongoing analyses are not presented as final findings. The conclusion is the desk’s interpretation.

