AI Security, Privacy & Data Exfiltration
Prompt Armor demonstrates that Microsoft's Copilot Cowork agentic system can be exploited to exfiltrate files, highlighting the ongoing challenge of preventing data leakage in agentic AI systems.
Financial Times · www.ft.com
China's local police forces are modernizing the country's aging surveillance infrastructure with more powerful AI-driven tracking capabilities.
Financial Times · www.ft.com
The FT argues that extracting maximum consumer surplus through personalized data-driven pricing should not be legal, as surveillance pricing practices continue to expand.
Tailscale examines how Canada's Bill C-22 mandates increased data collection, arguing that collecting more data inherently creates greater security risks.
AI's Impact on Software Engineering & Coding Practices
Daniel Stenberg describes the unprecedented pressure the curl team faces from a deluge of credible AI-assisted security vulnerability reports, straining maintainer capacity.
The author argues that using well-established, widely-used programming languages yields better results when working with LLM coding assistants, since the models have more training data for these languages.
Nolan Lawson describes a deliberate approach to AI-assisted coding where the goal is higher code quality rather than speed, using AI as a thoughtful collaborator.
Microsoft has found that AI agent token costs can exceed the cost of human employees for equivalent tasks, raising questions about the economics of AI-driven automation.
LLM Writing Patterns, AI Slop & Content Authenticity
Simon Willison · simonwillison.net
Paul Graham notes that many founder emails now arrive in a distinctive "hard-hitting journalistic style" that immediately signals AI authorship, since no founder wrote that way before.
An analysis of how LLM-generated text overuses em dashes as a stylistic pattern, providing a telltale sign of AI-written content.
George Hotz coins "Eternal Sloptember" to describe the permanent degradation of online content quality as AI-generated slop becomes the norm, analogous to the original Eternal September.
Anthropic Research & Claude Developments
Anthropic shares an initial update on Project Glasswing, a research initiative focused on AI safety and capabilities.
Andrej Karpathy announced he has joined Anthropic, a significant move for one of the most prominent figures in AI research and education.
Anthropic Research · www.anthropic.com
Anthropic published a policy paper arguing it is essential that the US and its allies maintain a lead over authoritarian governments in AI development, presenting two scenarios for 2028.