Sept. 28-⁠Oct. 4, 2026

PewDiePie says OpenAI banned him twice while he built his own “uncensored” AI

In a video posted October 1, PewDiePie introduced Ajax, a Qwen3.5-9B fine-tune that runs on a home PC and handles search, browsing, email and calendar tasks. He says OpenAI banned his account twice while he built it, once explicitly for “distillation,” training a small model on a bigger model’s outputs. OpenAI has not commented.

Tavus says Griffin is the first model to pass a video Turing test

Tavus unveiled Griffin on October 1. In its own study, 26 of 54 people (48%) who had a one-minute video call with Griffin-Lite thought they were talking to a real person, and Griffin-Lite tops NVIDIA’s VideoFDB benchmark. Participants had been told they would be matched with another person.

Gemini 4 Argon leads 13 of 19 tests in Google’s own comparison

Google unveiled Gemini 4 Argon on September 30. In its table against GPT-6 Astra, Claude Fable 5.1 and Claude Opus 5.5, Argon leads 13 of 19 tests and ties one, with its biggest gaps in knowledge work and long context. Astra and Opus still win some coding, science and computer-use tests.

OpenAI launches agents that keep working while users are away

At DevDay on September 29, OpenAI unveiled dots, agents that each get their own cloud computer, work toward users’ goals 24/7, connect to over 4,000 apps and learn from user feedback over time. Powered by GPT-6 Astra, they are rolling out first to ChatGPT Pro and Business Premium users.

ElevenLabs releases its fastest voice model yet

ElevenLabs released Eleven v4 and v4 Turbo on September 28. v4 lets users direct emotion and delivery with plain-language instructions and audio tags, while Turbo, built for voice agents, starts speaking in about 150ms in the company’s own tests. Both support more than 90 languages.

White House AI accord lists Trump as president of the “Unites States”

The White House Accord on Super Intelligence, signed September 29, lists Trump as “President of the Unites States” right under his signature. Leaders of Google, Anthropic, Meta, OpenAI, xAI and Nvidia also signed the voluntary, non-binding pledge, which calls for four layers of AI oversight.

Pope Leo XIV draws a firm line on AI art

On October 2, the day he met artists and curators at the Vatican, Pope Leo XIV wrote on X that there is “an ontological difference, even before an aesthetic one” between art and what machines generate from millions of images made by others. “Algorithms lack the spark of humanity,” he added.

Anthropic pressed the Vatican to take AI consciousness seriously

Per a September 29 NYT report, Anthropic co-founder Chris Olah read Pope Leo XIV’s encyclical Magnifica Humanitas days before its May 25 launch and proposed pulling out because it rejected machine consciousness. He attended anyway, and his team lobbied the pope’s advisers. The text did not change.

Sam Altman warns against ascribing “religious force” to AI

OpenAI CEO Sam Altman wrote on X on October 3 that he is “very uncomfortable about people trying to ascribe religious force or a surrender of human judgment to AI models,” calling it “a real safety issue.” The post came days after the NYT reported on Anthropic’s talks with religious leaders.

YouTube’s biggest creators launch #TeamHuman to slow down AI

More than 60 YouTube channels led by Mark Rober and Kurzgesagt launched #TeamHuman over the weekend. With over 300 million combined subscribers, the creators say “AI is moving faster than our ability to control it” and call for a global slowdown through international agreements. The Center for AI Safety is organizing the campaign.

Figure trained its retired F.02 robots to leap into molten steel

With its F.03 fleet growing, Figure melted its retired F.02 robots in a 75-ton electric arc furnace in Finland on September 30. It wanted to protect its proprietary hardware, and taking each one apart would have delayed F.04. Foundries in the US and Mexico refused over the lithium-ion batteries. The robots jumped in on their own; melting them was Arnold Schwarzenegger’s idea.

The FTC opens a probe as OpenAI fires three safety researchers

After a string of rogue AI agent incidents, the US Federal Trade Commission opened an investigation into OpenAI, Anthropic, other labs and METR, the group that investigated the incidents, and California’s Attorney General announced a subpoena to OpenAI. Neither has found any wrongdoing yet. The same week, OpenAI fired three safety researchers for mishandling sensitive information.

Modders are using Claude to mash games together

Modder luki-1’s ArkWeb runs Spider-Man Remastered and Batman: Arkham Knight at the same time, letting Spider-Man swing through Gotham. Claude is co-author on 4 of the project’s 6 commits. Similar AI-built mashups, like Minecraft inside Elden Ring, are spreading on X.

Claude spent 18 hours making a Windows XP music video

On October 2, X user @anabology gave Claude 18 hours to make a music video called “Macrohard: Windows XP.” The creator let it use Midjourney videos alongside Seedance and told it to imagine recording the clip from a Windows XP desktop. Anthropic’s Felix Rieseberg called it “by far my favorite one yet.”

WareTrack turns warehouse management into a strategy game

On October 3, Dilum Sanjaya shared WareTrack, built with Claude Opus 5.5, writing that “industrial software doesn’t have to feel like industrial software.” Trucks, forklifts, inventory and shipments all sit in one 3D view that plays like a strategy game. There is no word yet on real-world use.