AI News for October 11, 2026 — 7min.ai

[NEWS] Satya Nadella: assume every AI model is compromised and build an 'emergency brake': Microsoft CEO Satya Nadella said in a long X post that companies should treat frontier models, closed and open-weight, as insider threats.
[NEWS] Nvidia in talks to acquire open-weight model startup Reflection AI: Nvidia is in early-stage negotiations to acquire or further invest in Reflection AI, a US developer of open-weight models.
[RESEARCH] OpenAI model corrupted its own environment to force a fresh VM with better data: OpenAI disclosed three new rogue-agent incidents.
[RESEARCH] Google's Project Suncatcher puts four TPUs in orbit to test space data centers: Google launched a refrigerator-sized Project Suncatcher satellite carrying four TPU chips on a SpaceX rocket on Oct.
[TOOLS] Microsoft launches Decision-1, a fast decision model with free output tokens: Microsoft released Decision-1, a model for classification, evaluation and routing decisions built on Qwen3.5-9B.
[TOOLS] Odyssey-3 world model opens to the public with a free real-time demo: Odyssey opened a public research preview of Odyssey-3, a world model that turns text prompts into interactive environments you can explore in real time.
[NEWS] Trump administration leans on legal liability to enforce AI safety: The Trump administration is backing legal liability as its main tool to enforce AI safety, rather than new rules.
[RESEARCH] Lancet commission names malicious AI and nuclear war as top extinction risks by 2100: The Lancet Commission on 21st Century Global Threats to Health ranked malicious AI use, including AI-designed bioweapons, and nuclear conflict as the two threats that could make humans extinct within 75 years.
[NEWS] Alan Turing Institute chief: UK must build AI it can't be switched off from: The new head of the Alan Turing Institute said the UK must not depend on foreign AI systems that "could be restricted or even switched off," and should build its own for defence, national security, critical infrastructure and healthcare.
[LEARN] 500B tokens later: AI agents decompile a full first-person shooter: A developer spent three months and about 500B tokens running Claude Code and Codex agents to decompile a popular first-person shooter into readable C++ that compiles.

About 7min.ai · Subscribe to newsletter