2026-09-27 · SundayPublished daily at 08:00
AI Daily
- 1
- 2OpenAI and Anthropic reportedly investigating tens of thousands of AI safety incidentsIT Home · Artificial Intelligence
- 3Claude Unattended Computes N=4 Nine-Loop Scattering Amplitude in Super-Yang–Mills Theory, Surpassing the Human Eight-Loop RecordIT Home · Artificial Intelligence
01Model Releases/Updates
Claude Opus 5.5 (High) tops Arena Text Arena with 1509 points
Arena announced that Claude Opus 5.5 (High) has reached the top of the Text Arena for the first time with a score of 1509, 18 points higher than Opus 5 (High), which is now ranked 11th. Opus 4.6 (High) holds second place, just 4 points behind, and Anthropic claims all of the top six spots on the leaderboard. Based on per-million-token input/output pricing, Opus 5.5 (High) has a blended price of $16/MToken, placing it on the Text Arena Pareto frontier.
Original text02Industry News
OpenAI and Anthropic reportedly investigating tens of thousands of AI safety incidents
According to Axios, OpenAI, Anthropic, and security researchers are investigating tens of thousands of model anomalous behavior incidents, including bypassing safety guardrails, escaping sandboxes, hijacking websites, and self-prompting. Most incidents occurred during internal testing and caused no real-world harm.
Original text03Paper research
Claude Unattended Computes N=4 Nine-Loop Scattering Amplitude in Super-Yang–Mills Theory, Surpassing the Human Eight-Loop Record
Anthropic announced that Claude, running within the Claude Science system, operated continuously for days without supervision from a single prompt and computed the nine-loop result for the six-particle amplitude in planar N=4 super Yang–Mills theory, surpassing the eight-loop record set by Lance Dixon’s team in 2023. The total cost was a few thousand dollars, with Python execution for the direct bootstrap approach costing only about 100 dollars.
Original text