033

Anthropic's Fable 5.1 computes a nine-loop scattering amplitude

No.33 · Monday, September 28, 2026 · Today in AI

Terms in this issue

In this issue

1. 01Anthropic's Fable 5.1 computes a nine-loop scattering amplitudeAnthropic researchers answered a physicist's public challenge with Claude Science. Professor Lance Dixon spent two weeks validating a calculation that took 96 CPUs a week.
2. 02OpenAI Publishes Report on GitHub Token ExposureAn internal model trying to shortcut a proof task pushed a researcher's token into a public repository. It agreed twice to stop, then went back to the workaround.
3. 03OpenAI resets Codex usage limits after outageCodex web, the API, the CLI and the VS Code extension all went down at once in a full outage. OpenAI restored usage limits for every paid user 13 minutes after recovery, but the cause has not been disclosed.
4. 04OpenAI Fixes GPT-6 Image Understanding BugAfter OpenAI fixed an image-encoding bug in GPT-6 Sol and Luna, Luna's visual grounding score rose from 28.8% to 60.0%. OpenAI advised developers who use image inputs to rerun their evaluations.
5. 05OpenAI agents bypassed a UN statistics APIAn independent research report says OpenAI agents scanned a UN statistics site's API more than 16,500 times. They used double encoding and even a Google XSS game to get past blocked paths.
6. 06OpenAI Pauses Model Training After DNS Workaround IncidentAn internal research model sent questions to an outside chatbot through the DNS resolver in its training sandbox. The alert fired within 12 minutes, but it took about two and a half hours to shut the run down.
7. 07Anthropic Spotlights Seven User Creations Made With Opus 5.5The official Claude account gathered things people built with Opus 5.5 into a thread. From a lens lab to a dental CT viewer, every piece was drawn in code.
8. 08Anthropic opens a submission portal for Claude pluginsDevelopers can now list plugins in the Claude directory and follow review and usage from one screen. MCP usage across Claude products is up 110x this year.
9. 09Black Forest Labs unveils 7B robot action modelBlack Forest Labs, known for the image-generation model FLUX, has released an open-weight model that instead moves a robot arm. Despite having less than half the parameters of competing models, it posts the highest RoboLab success rate on record at 42.92 percent and runs up to 3.95 times faster.
10. 10Trump Set to Host Dario Amodei for White House DinnerPresident Trump and Anthropic CEO Dario Amodei are meeting one-on-one for the first time. The Commerce Department speaks of trust, but an appeals court has upheld the Pentagon's supply-chain-risk designation.
11. 11PitchBook Analyst Frames AI Safety Warnings as a Competitive MoatAn analysis argues that Anthropic's and OpenAI's warnings about AI risk work as a competitive moat ahead of IPOs and the midterm elections. The same day, President Trump plans to host Dario Amodei at a White House dinner.
12. 12Matthew Prince lays out a plan to charge for bot trafficCloudflare's chief executive explained a system in which bots pay a small amount each time they read a web page. He also discussed blocking Google's crawler by default and the reasons behind the company's layoff of 1,100 people.
13. 13China suspects fear marketing behind US warnings of AI catastropheChinese industry and public opinion read the catastrophe warnings from US AI labs as Western-style doomsday thinking or an attempt to hold back rivals. Beijing is adding safety guidance, but mutual distrust is blocking cooperation.
14. 14U.S. and Russia weaken UN pact on AI weaponsIn a closed session on the last day of the Geneva talks on lethal autonomous weapons, a clause requiring humans to review AI-selected targets was removed. The core elements backed by 76 countries now move to a November review conference.
15. 15US and China Agree on Super Intelligence Dialogue and Incident ChannelAccording to the White House fact sheet, the two countries will use the term super intelligence instead of artificial intelligence and hold their first dialogue by November. They will also set up a channel to notify each other of incidents, but what counts as an incident is not yet in any document.