Maximilian Schreiner
Max is the managing editor of THE DECODER, bringing his background in philosophy to explore questions of consciousness and whether machines truly think or just pretend to.
OpenAI releases open-source model that strips personal data from text
OpenAI has released Privacy Filter, an open-source model designed to detect and redact personal data in text.
Read full article about: Google says 75 percent of its new code is now written by AI
75 percent of new code at Google is now generated by AI and then reviewed by human developers, the company says. That number has climbed fast: it was 25 percent in October 2024 and 50 percent by fall 2025. In a blog post, Google CEO Sundar Pichai said the company is now shifting to "agentic workflows" where AI systems operate with increasing autonomy.
A complex code migration was completed six times faster than a year earlier thanks to collaboration between AI agents and developers. Google's engineers use the company's own Gemini models for these tasks.
Some employees at Google DeepMind are also allowed to use Anthropic's Claude Code, according to a report by Business Insider - likely because Google hasn't built a real alternative yet. A dedicated team within Google Deepmind is now working to close the gap with Anthropic on AI coding tools.
Read full article about: Sony AI builds the first robot to reach expert level in a sport
Sony's table tennis robot "Ace" is the first robot to reach expert-level performance in a sport, according to the company. In matches played under the official rules of the International Table Tennis Federation in April 2025, Ace beat three out of five elite-level players but lost to professional players. As of December 2025, Ace is winning against pros, too.
The robot uses nine cameras, three vision systems, and AI-powered controls to track the ball and its spin faster than the human eye can. Professional player Mayuka Taira said the robot is hard to read because it shows no emotion. According to project lead Peter Dürr of Sony AI Zurich, the underlying technology could also be applied in manufacturing and service robotics.
Table tennis expert Kinjiro Nakamura, who competed in the 1992 Olympics, said after watching one of Ace's shots that he wouldn't have thought the shot was possible - but if a robot can pull it off, maybe humans can, too.
Read full article about: Unauthorized users breach Anthropic's restricted Mythos AI model
A small group of unauthorized users gained access to Anthropic's new AI model Claude Mythos, Bloomberg reports. Anthropic considers Mythos powerful enough to enable dangerous cyberattacks, which is why the company only makes it available to select partners like Apple, Amazon, and Cisco through its "Project Glasswing" program.
The users, members of a private Discord channel, got in on the day of the announcement. They pulled it off using the access credentials of a member who works as a contractor for Anthropic, along with publicly available information from a data leak at AI startup Mercor. According to Bloomberg, the group didn't use Mythos for cyberattacks but for harmless tasks like building simple websites for testing.
The source says the group also has access to a number of other unreleased Anthropic AI models. The company says it's investigating the incident. So far, there's no indication that the access extended beyond the external contractor's environment or that Anthropic's own systems were compromised.
Read full article about: Meta tracks US employees' clicks and keystrokes to train AI agents
Meta is installing new surveillance software on its US employees' computers that captures mouse movements, clicks, and keystrokes to train AI models. That's according to Reuters, citing internal memos.
The tool, called Model Capability Initiative (MCI), runs on work-related apps and websites and occasionally takes screenshots. The goal is to build AI agents that can handle work tasks on their own, like navigating dropdown menus or using keyboard shortcuts. Meta spokesperson Andy Stone said the data won't be used for performance reviews and that sensitive content is protected.
As part of the Agent Transformation Accelerator initiative, CTO Andrew Bosworth announced that agents will eventually take over the bulk of the work. Meta plans to cut ten percent of its global workforce starting May 20. Legal experts like Valerio De Stefano of York University warn that these practices would likely run afoul of GDPR in Europe.
Musk's SpaceX bets $60 billion on Cursor to fix xAI's coding gap
SpaceX has secured an option to buy AI coding startup Cursor for $60 billion. The deal would fill a gap Elon Musk’s xAI has struggled to close on its own: coding tools.
