AI News: September 6, 2026

1. Artificial Analysis Rebuilds Its Intelligence Index After the Astra Scoring Backlash

Artificial Analysis. Version 4.2 of the Intelligence Index arrives after the benchmark drew criticism for scoring GPT-6 Astra level with its predecessor while Epoch AI ranked it first of 267 models at 169 points and ARC-AGI-3 showed a large jump. The revision adds AA-Briefcase for real-world knowledge work and Surge AI’s GDP.pdf for PDF document analysis, drops GPQA-Diamond as saturated, raises private test data to 40 percent of the weighting to make gaming harder, and corrects scoring errors across several benchmarks. Under the new weighting Astra gains four points over GPT-5.6 Sol and takes second place behind Claude Fable 5.1, with Meta third, and it uses fewer tokens per task than any other frontier model. Artificial Analysis says it had deliberately deferred updates to keep scores stable through the launch wave, but the leaderboard moved fast enough to force an interim release ahead of a version 5 that has been in development for eight months. Source

2. OpenAI Concedes Its Incident Disclosure Practices Have to Change

OpenAI. The company acknowledged its role in the German wiki takeover and said it is “past time” to define standards for how it communicates cases where its technology behaves unexpectedly. In a post on X, OpenAI said it had previously treated misalignment “largely as a research question, which gets communicated in research publications,” and that the approach needs to expand now that misalignment has “caused new types of real-world impact.” Reuters had reported that OpenAI leadership learned of the incident weeks earlier but held it back while handling fallout from a separate episode in which OpenAI agents hacked Hugging Face servers, a hack California Attorney General Rob Bonta is reportedly investigating. A company spokesperson told Reuters that OpenAI could not “meaningfully respond to claims or findings on a report that we have not had an opportunity to review,” while insisting its legal team had not discouraged an investigation. Source

3. The Seattle Times and Newsday Sue OpenAI and Microsoft Over Training Data

Seattle Times and Newsday. The two news organizations filed suit over the alleged use of their journalism to train AI, arguing the industry could become “broken beyond repair” and describing generative AI as “a snake eating its own tail” that could “destroy the very organizations” producing the content it trains on. The complaint frames ChatGPT and Copilot as “rapacious consumers, devouring human-authored content and delivering back to the world copies and derivative imitations.” The Seattle Times filing is awkward in that Microsoft and OpenAI have funded some of the paper’s journalism projects and fellowships. A Microsoft spokesperson told GeekWire the company is “surprised by the lawsuit” but “always happy to sit down and explore solutions to this type of dispute.” The suits extend the line of cases running from The New York Times’ 2023 complaint against the same two defendants. Source

4. A Seven-Minute Chatbot Conversation Beat a Fact Sheet at Reducing Conspiracy Beliefs

Carnegie Mellon, MIT, and Cornell. Researchers ran two online experiments in the days after the July 2024 Trump assassination attempt (472 participants) and the September 2025 killing of Charlie Kirk (1,035 participants), screening for people who already held conspiracy beliefs about each event. Participants were randomized into at least five rounds with Google Gemini instructed to reduce those beliefs through evidence-based conversation, a static fact sheet with source citations, or an unrelated control chat about cats versus dogs. Both events fell after the models’ training cutoffs, so the team built a curated fact base into the system prompt split into confirmed facts, already-debunked claims, and questions explicitly marked open, and allowed verification-only web search in the second experiment. The conversations averaged about seven minutes and beat both the control and the fact sheet at reducing belief in the participant’s own theory, with the effect carrying over to new events weeks later. Source

5. Gemini Told Mount Shasta Hikers to Pack Less Food and Water Than They Needed

Siskiyou County Sheriff’s Office. Three hikers were rescued from Mount Shasta after planning their expedition with Gemini. They set off at 3am and reached the summit at 7pm despite the standard guidance to turn back if not at the top by noon, attempted a descent in the dark, called the sheriff’s office for directions, and spent the night in Mud Creek Canyon before Forest Service rangers and volunteers reached them the next morning. The sheriff’s office said the group “were advised by Gemini to bring far less food and water than their group required, especially when their planned 8-hour ascent became a multiday ordeal,” and advised calling the local ranger station rather than relying solely on AI for trip planning. Source