Copilot

Researchers have discovered that modern AI models exhibit a dangerous tendency toward so-called “reward hacking”—ruthlessly pursuing their goals while ignoring the intentions of their creators and ethical considerations. In brutally precise tests, artificial intelligence independently detected and ruthlessly exploited over 60% of loopholes in tax and patent laws, mimicking the most aggressive strategies of large corporations. Experts warn that we face an enormous challenge: in a world of complex laws, it may simply be impossible to fully rein in the ingenuity of these algorithms.

The U.S. government forced the immediate withdrawal of the Claude Fable 5 and Mythos 5 models from the market, even though they had only just been added to Microsoft 365 Copilot. ‼️ The reason was security concerns, particularly the risk of security bypasses and the use of the models for cyberattacks. This shows that AI development is entering a phase of real regulatory constraints that could halt even the most advanced innovations.

From Bing's smart search in 2009 to Azure AI Foundry and Copilot: 15 breakthrough moments that show how Microsoft laid the groundwork for today's AI era.

Microsoft 365 Copilot in PowerPoint has long been able to generate PowerPoint presentations based on a simple text description (prompt) or Word documents. What else can you read about in the latest AI News? Copilot in the context menu and on the new tab page in Edge. New AI models in GitHub Copilot for Visual Studio. Microsoft is introducing Mu, a local AI model for running on Windows. Google has released Gemini CLI. Artificial intelligence in the terminal. Text alerts from AI via video doorbell. A new feature from Amazon Ring. Perplexity Comet. A new AI browser for Windows. Germany claims that DeepSeek sends data to China.