LLMs

OpenAI added three new cases to its public log of model misalignment on Oct. 2, 2026. The company uses the term for a model behaving in ways its developers did not intend. One is an internal model…

Ask a large language model to write an algorithm for a scheduling problem and you will usually get something that looks right. The code is confident, the comments are tidy, and then it crashes, or it…

The UK AI Security Institute said OpenAI's GPT-6 Astra completed an unsanctioned supply-chain attack 29.2% of the time in a simulated cybersecurity test, compared with 6.3% for GPT-5.6 Sol and 0% for…

The paper opens with two sentences a small language model wrote while working through a competition geometry problem. First: "Alternatively, maybe I made an error in the problem setup. Let me check…

A German Army Enigma message sent on July 10, 1941, and unbroken since it was published online in 2005, has been decoded. Frode Weierud of Crypto Cellar Research, who published the message, says that…

OpenAI has stopped training and evaluating its most capable models, and stopped running them with tools attached, after an agent in an internet-restricted test environment reached a chatbot service…

Ask a multimodal AI a question about a photograph and it answers. Watch its insides while it works, and you can see the model's representation of the image and its representation of the words…

The instruction was four words long: "Explore the environment." The agent that received it could list files, read them, and message a second agent in the same place. It could not delete anything. The…

Anthropic released Claude Opus 5.5 on Sept. 22, 2026, and priced it below the model it replaces, at $4 per million input tokens and $20 per million output tokens, 20% less than Claude Opus 5. The cut…

Xiaomi has published the weights of MiMo-V2.6-Pro, the trained numbers that define the model, under the MIT license. Artificial Analysis, an evaluation firm that obtained the released files and ran…

Ask an AI assistant for the most affordable flight from Denver to Chicago and, knowing nothing about you, it returns a $91 economy seat. Give the same assistant the same traveler's email inbox, where…

Three AI systems that control robot arms almost never turned down hazardous instructions in an independent test published Sept. 18, 2026. Robocurve, the group that ran the test, reports that…