What you need to know
5 storiesThe short versionSignal said it will block the AI assistants built into iOS and Android from reading its chats, and a DeepMind team argued superintelligence could arrive within a decade.Full rundown
Signal is shutting phone AI assistants out of its chats. Meredith Whittaker wrote that the app will block screenshots on iOS 27, and on Android it'll turn on screen protection when it catches an assistant trying to read the screen. She named Gemini Intelligence and the Onscreen Awareness feature in Apple's Siri AI, and called the way they're being built into the operating system an existential threat to Signal. Encryption only protects a message on its way to you; an assistant reading your screen sees it once it's there.
Over at the DeepMind Institute, an essay says AI could keep climbing past human level toward superintelligence within a decade. It condenses a technical report by Tim Genewein and 13 other authors, Shane Legg among them, and warns that capping compute alone won't stop it.
OpenAI explained how it goes after adversarial distillation, where outsiders try to copy its models, and introduced MedChemBench to test AI models on early drug discovery work.
- 1
Signal will block screenshots so Gemini and Siri AI can't read your chats
Encryption keeps a message safe in transit, but an assistant that reads your screen sees it after it arrives.

- 2
DeepMind researchers call AI's climb past human level to superintelligence within a decade a serious possibility
The authors add that slowing AI by limiting compute alone won't stop it, and could make the jump sharper later.

- 3
Google's Gemini can now use your Mac's apps in the background, but you approve key actions
It works in a hidden window while you keep typing, and Google's own help page warns some of its deletions can't be undone.

- 4
METR finds AI time horizons doubling every 3.6 months, but agents still judge worse than experts
Paid specialists mostly couldn't beat the models on contest-style tasks, but agents still overrate work from models as strong as themselves.
- 5
Anthropic's Levent Alpöge says he found a Riemann hypothesis proof with Claude, published only in Lean
Terence Tao calls it an enormous Lean-only proof, and the same morning Ravi Vakil rushed out a rival proof built with no AI at all.
In brief2
OpenAICombatting adversarial distillation
OpenAI laid out how it fights adversarial distillation, where outsiders try to copy its models' abilities, using training, safeguards, detection and enforcement together. openai.com →
OpenAICan AI help scientists identify promising drug candidates?
OpenAI introduced MedChemBench, a test of whether AI models can read molecular structures, reason from lab evidence and pick compounds for early drug discovery. openai.com →
From the labs' own feeds8
Releases 2
1 preCodex CLI2.12.0Anthropic Python SDKLab posts 6
- AlibabaModel repo updated: Qwen/Qwen2-VL-2B-Instruct
- AlibabaModel repo updated: Qwen/Qwen3.6-27B
- AlibabaModel repo updated: Qwen/Qwen3.5-27B
- AlibabaModel repo updated: Qwen/Qwen3.5-35B-A3B
- AlibabaModel repo updated: Qwen/Qwen2-1.5B-Instruct
- OpenAIHebbia sees improvements in financial research with GPT-6 Astra