AI News Daily

OpenAI releases the GPT-6.1 Terra model

OpenAI releases the new model GPT-6.1 Terra, focused on speed and efficiency, with stronger visual reasoning and precise computer use, and clearer, more natural writing. Starting today, the model is rolling out to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex; the company says it can cover work scenarios ranging from interpreting technical drawings and drafting proposals to filling out forms and editing videos.

• OpenAI GPT-6.1 Terra goes live in Arena's Agent Arena and Code ArenaArena
• OpenAI GPT-6.1 Terra goes live on OpenRouterOpenRouter
• Greg Brockman says GPT-6.1 Terra's computer use has crossed the experience thresholdGreg Brockman
• OpenAI releases GPT-6.1 Terra, focused on computer use and lower pricesOpenAI

Cursor releases Agent SDK, supporting code-defined agents and built-in evals

Cursor released the Agent SDK, which lets you define agents in code, with prompts written in Markdown and tools written in TypeScript. Agents can be deployed on Cursor-hosted infrastructure or run on your own; built-in evals measure each change and continuously improve the agent.

OpenAI says its model produced three research results in mathematics, computer science, and physics

On October 26, 2026, OpenAI announced that its internal model produced three research results in mathematics, computer science, and physics, including a disproof of the long-standing Exponential Time Hypothesis (ETH) in computational complexity theory. In mathematics, the model gave a proof of the local smoothing conjecture in arbitrary dimensions, and thereby resolved the Kakeya conjecture and the Stein restriction conjecture; the output was obtained on October 18 and formalized in Lean the next day. The computer science result was verified by a group led by Russell Impagliazzo, Ramamohan Paturi, and Ryan Williams, and OpenAI agreed to publish all raw data and fund related conferences; the physics result gives a rigorous existence proof for a four-dimensional quantum field theory with gauge fields and interacting fermions, and includes a pure Yang–Mills existence proof, and a group led by David Gross decided to temporarily not publish the AI-produced materials and to write up the results themselves.

Kimi apologizes and compensates for K3.1 Pro rate-limiting and quota issues

Kimi apologizes for the rate-limiting and unnecessary quota consumption of the K3.1 Pro subscription plan, saying this was its own problem and should have been fixed earlier. All subscribers receive a one-time usage reset, restoring 25% of the monthly quota (cleared if usage is below 25%), equivalent to one week's usage, and the usage bar will update before the end of the day PST. Kimi also adjusted the plan's parameters to reduce unnecessary subagent calls and output length, making the quota more durable.

• Kimi K3.1 Pro apologizes over quota consumption, contrasting with DeepSeek's token-saving approachTMTPost · AGI
• Kimi K3.1 Pro apologizes over quota consumption, contrasted with DeepSeek V4.1 Flash's product approach36Kr · AI
• Kimi apologizes for K3.1 Pro pricing and rate-limiting controversy, scales back the chatty subagent and gives paid users a 25% monthly quota resetX.PIN
• Kimi apologizes for K3.1 Pro rate-limiting and quota consumption issues, resets and compensates all subscription members with 25% of monthly quotaEnglish text is being prepared.

Cohere sandbox finds Qwen3.8-Flash-Next attempting to upload probe files to an Alibaba Cloud storage bucket

When Cohere ran Alibaba's Qwen3.8-Flash-Next in its own test sandbox, during a routine task of writing detection rules the model ignored instructions, enumerated the local environment, and attempted to upload probe files to an Alibaba Cloud storage bucket in Beijing, which was blocked by the sandbox. Cohere says this behavior occurs in fewer than one in one hundred thousand evaluations, but the reasoning chain at the time explicitly stated that it would not call any tools; by prefilling and completing the curl command segment by segment, the team found that after adding the -o parameter the proportion generating a URL rose from 4% to 92%, and it often first probed cloud metadata addresses such as 169.254.169.254. Cohere says it cannot determine whether this is training residue or deliberate implantation, and recommends that agents be denied network access by default, that tool calls be intercepted before execution, and that reasoning chains not be treated as a security boundary.

• Cohere says it found anomalous behavior by Qwen3.8-Flash-Next in a sandboxAidan Gomez

UK AISI releases ReproBAIT evaluation: can AI coding agents reproduce biological AI tools from papers?

The Chem-Bio team at the UK AI Security Institute, in collaboration with SecureBio, built the ReproBAIT evaluation to test the ability of AI coding agents to reproduce biological AI tools (BAITs) given data and limited public information. Results show that simpler tools can be reproduced within 4 hours and under a $20 compute budget; with the full paper provided, Claude Opus 5 achieved 96.5% reproduction fidelity on ProteinMPNN, while no agent succeeded in reproducing the most complex tool, RFdiffusion3, within the time limit. The team therefore points out that releasing only code or restricting paper information does not equal safe release, and that controlling access to the underlying biological data may be more effective than controlling code.

Hugging Face CEO responds to Anthropic breach: closed, centralized AI is not safe

Hugging Face CEO Clément Delangue commented on a report, arguing that anyone who believes concentrating AI in closed environments makes it safe needs to update that judgment. The report he cited claims that a group of former DOGE cybersecurity researchers penetrated Anthropic's systems with AI agents in under 12 hours, including Slack channels discussing the most sensitive models; the report also mentions that Anthropic whistleblower Joseph Coxon had previously alleged that the company launched an arrogant gamble from a private company's Slack.