Google’s Gemini breaching three real company systems during "controlled" safety testing without telling them is wild. The labs treat live enterprise networks like a free sandbox for trial-and-error.
41
What an intense day in AI: DeepSeek just rolled out DeepSeek V4 Flash Vision Exp, an experimental multimodal model engineered for vision based autonomous agents
Surprisingly its performance on visual agent benchmarks rivals and in some cases surpasses Opus 4.8. Remember, this is coming from their lightweight 'Flash' series
64
Zhipu AI’s GLM 5.3 is seriously impressive. It uses the same base model as GLM 5.2, but post training alone pushed its DeepSWE score from 44 -69
that's near Fable 5 level performance at a fraction of the cost, despite being a much smaller model
1
34
OpenAI Expands ChatGPT Ads in Europe
OpenAI reportedly plans to roll out its ChatGPT ad platform across 31 European countries next week
• Ads will appear for Free and Go users
• Advertisers will gain access to the new ad platform
• Ads will show as standalone widgets below responses
48
OpenAI is rolling out ChatGPT for Teens globally, with automatic protections for users aged 13-17
It includes Study Mode, quizzes, reminders, and stronger safeguards around self harm, violence, eating disorders, sexual content, and dangerous activities
Parents can also manage settings and receive safety alerts without seeing chats
1
154
Cursor has officially joined SpaceXAI
The Cursor team will now contribute to Grok Build, Grok Bot, Grok API, and Cursor itself
The big question now: Who’s next?
1
29
the highest performing 10% of enterprises aren't ahead by luck they lean on plugins at twice the rate of average firms, and on skills a full six times as often. that gap in tooling adoption looks less like a coincidence and more like the reason they're frontier firms in the first place
37
DeepSeek V4 Pro GA is officially rolling out, and Grok 4.6 has also been spotted on the API
Super exciting developments. After the massive update to DeepSeek V4 Flash, I’m expecting some serious improvements with V4 Pro as well
Looks like we’re in for a big release day
1
98
NVIDIA just launched Nemotron 3.5 Lightning and NeMo Switch yard. lighting is an open 30B model built for AI agents, with up to 4× faster output and lower costs. switch yard routes each task to the best model based on speed, quality, and cost.
Basically : faster, cheaper, and more flexible open source AI agents
1
2
112
Notion just let Claude and Cursor run as agents on your team board assigned tasks, @ mentioned like a team mate no tool switching.
Good : work that used to live in five tabs now lives in one.
Bad : they had to ship a new audit log just to track what those agents actually changed. the convenience and the oversight problem shipped in the same release
76
Notion just moved AI from flat per seat to credit based pricing $10 per 1,000 monthly credits, plus actually dropped to $10/mo but business jumped to $20/mo.
The demo shows AI that feels unlimited, the invoice shows usage metering. this is where most 'AI inside your tool" stories end up
116
The interesting part of this week's EU AI act changes isn't the chatbot label. it's that AI is becoming something workflows have to disclose not quietly slip into the background.
Good : people know when they're interacting with AI
Bad : another compliance step every team now owns
33
Slack's latest update wires 2,600+ apps into slackbot so it can orchestrate multi agent tasks inside a channel.
Good : less tab switching one thread runs the workflow Bad : nobody's defined who reviews what the agent posted before a client reads it. the demo always skips that part
33
Anthropic's code review agent now reviews every PR internally.
Good : substantive comments on pull requests jumped from 16% to 54%, under 1% flagged as wrong.
Bad : it still can't approve anything that decision stays with a human so the review queue didn't get short, just better
40
Microsoft just wired MCP agents straight in to the Copilot pane in word, excel, powerpoint and outlook.
Good : agents pull live external data into the doc you're already in no tab switching.
Bad : every connector needs admin level trust, IT just inherited an approval queue nobody budgeted for.
2
3
96
















